Computers and Software

International Review on

Computers and Software (IRECOS) Contents ISPM: Improved Snow Prediction Model to Nowcast Snow/No-Snow by Kishor Kumar Reddy C., Vijaya Babu B.

536

Bandwidth Allocation in Wireless Mesh Network Using Efficient Path Selection Scheme by Karunya Rathan

545

A New Low Complexity Transform for Peak Power Reduction in Discrete Hartley Transform Based Multicarrier OFDM System by N. Bharathi Raja, N. Gangatharan

552

Agile Development with 3D Data Modeling by Kristian Sestak, Zdenek Havlice

558

An Advanced Approach for Extraction of Brain Tumor from Digital MR Images Using Morphology Gradient & ART by Abderrahim Khatabi, Amal Tmiri, Ahmed Serhir

566

Lloyd and Minkowski Based K-Means Clustering for Effective Diagnosis of Heart Disease and Stroke by D. Nalini, R. Periasamy

573

Chameleon Clustering Algorithm with Semantic Analysis Algorithm for Efficient Web Usage Mining by Anupama Prasanth, M. Hemalatha

580

Wavelet Processing for Neural Network Training Applied to Solar Radiation Forecasting by Andrés Ussa Caycedo, Robinson Jiménez Moreno, Darío Amaya Hurtado

586

Educational Data Mining: an Intelligent System to Predict Student Graduation AGPA by Z. Shana, S. Abdulla

593

Heraclitean Ontology for Specifying Systems by Sabah Al-Fedaghi

602

Adapted Bat Algorithm for Capacitated Vehicle Routing Problem by Anass Taha, Mohamed Hachimi, Ali Moudden

610

Web Cache Replacement with the Repairable LRU Algorithm by S. Prapai, P. Somchai, P. Nalinpat

620

Arabic Named Entity Recognition for Crime Documents Using Classifiers Combination by Suhad Al-Shoukry, Nazlia Omar

628

A Study on Data Collection Behaviors for a Class of Multi-Hop Wireless Sensor Networks Based on Probability Model by K. Jaksukam, S. Vorapojpisut

635

Embedding Image ROI Watermark Into Median DCT Coefficients by Nadhir Ben Halima, Osama Hosam

643

Errata corrige

535

.

Copyright © 2015 Praise Worthy Prize S.r.l. - All rights reserved

International Review on Computers and Software (I.RE.CO.S.), Vol. 10, N. 6 ISSN 1828-6003 June 2015

ISPM: Improved Snow Prediction Model to Nowcast Snow/No-Snow Kishor Kumar Reddy C.1, Vijaya Babu B.2 Abstract – Till date, many of the practioners, meteorologists, researchers, academicians, scientists across the globe proposed many methodologies and tools to nowcast snow/no-snow using satellite imagery, radar imagery, physical instruments, various algorithms, models and so on, adding to it some researchers estimated the amount of snow while some researchers detected the density of snow and few discriminated the differences between wet snow and dry snow. The main crux of the present research is to nowcast the presence of snow/no-snow more accurately by making use of historical weather datasets by adopting decision trees. In this paper, we are proposing a new algorithm Improved Snow Prediction Model (ISPM), an improvement to our earlier algorithms Snow Prediction Model (SPM), Improved Supervised Learning in Quest (ISLIQ), Supervised Learning using Gain Ratio as Attribute Selection Measure (SLGAS) and Supervised Learning using Entropy as Attribute Selection Measure (SLEAS). The ISPM algorithm out performs in terms of various performance measures like sensitivity, specificity, precision, dice, error rate and accuracy when compared with other decision tree models. The proposed method provides less computational complexity by evaluating the interval range, which significantly decreases the number of split points. Experimental results show that the ISPM algorithm scales up well to both large and small datasets with large number of attributes and class labels. Copyright © 2015 Praise Worthy Prize S.r.l. - All rights reserved.

Keywords: Decision Tree, SLIQ, SPM, SLGAS, SLEAS, ISLIQ, ISPM, Snow, No-Snow

I.

A similar brightening effect occurs when no snow is falling and there is a full moon and a large amount of snow [2]-[4]. The main crux of the present research is to nowcast the presence of snow/no-snow more accurately. Till date, many of the researchers, academicians, scientists across the globe proposed many methodologies in the prediction of snow using satellite imagery, radar imagery, physical instruments, various algorithms, models and so on, adding to it some researchers estimated the amount of snow while some researchers detected the density of snow and few discriminated the differences between wet snow and dry snow [2] [5]-[25]. Earlier, weather predictions were mainly based upon changes in current weather conditions, barometric pressure, and the condition of the sky [3]. In 650 BC, the Babylonians predicted the weather based on cloud patterns and also from astrology. Later, in about 340 BC, Aristotle described weather patterns in Meteorologica [3]. Later, Theophrastus compiled a book on forecasting of weather, called the Book of Signs [3]. By 300 B.C., Chinese astronomers developed a calendar, in which they divided a year into 24 festivals and each festival is allied with a different type of weather [3]. In 904 AD, IbnWahshiyya's Nabatean discussed about the forecasting of weather from changes in the atmosphere and also signs from the planetary astral alterations; signs of rain based on observation of the

Introduction

Precipitation occurs when a segment of the atmosphere becomes drenched with water vapour, so that the water condenses and precipitates. Snow is precipitation in the form of flakes of crystalline water ice that falls from clouds [1]. Since snow is composed of small ice particles, it is a granular material. It has an open and therefore soft, white, and fluffy structure, unless subjected to external pressure. Snowflakes come in a variety of sizes and shapes. Types that fall in the form of a ball due to melting and refreezing, rather than a flake, are known as hail, ice pellets or snow grains [1]. Fresh snow reflects 90% or more of ultraviolet radiation, which causes snow blindness, also reducing absorption of sunlight by the ground. Snow blindness is a painful eye condition, caused by exposure of unprotected eyes to the ultraviolet rays in bright sunlight reflected from snow or ice. This condition is a problem in Polar Regions and at high altitudes, as with every 300 meters (980 ft) of elevation (above sea level), the intensity of UV rays increases by 4%. Snow's large reflection of light makes night skies much brighter, since reflected light is directed back up into the sky. However, when there is also cloud cover, light is then reflected back to the ground. This greatly amplifies light emitted from city lights, causing the 'bright night' effect.


536

Kishor Kumar Reddy C., Vijaya Babu B.

lunar phases; and weather forecasts based on the movement of winds [3]. All the ancient weather forecasting methods, usually based on observed patterns of events. For instance, if the sunset was red, the particular day often brought fair weather. However, not all of these predictions on weather prove reliable, and many of the methods found that not accurate. This increased the interest in weather forecasting to most of the academicians, researchers, scientists and so on. Now, weather forecasting relies on computer-based models, which takes various atmospheric factors such as Humidity, Temperature, Pressure, Wind-Speed, DewPoint, Visibility, and so on into account [1]. Human input is still required to pick the best possible forecast model to base the forecast upon various statistical methods. The disordered nature of the atmosphere, the enormous computational power required to compute the equations that describe the atmosphere and an incomplete understanding of atmospheric processes mean that forecasts become less accurate as the difference in current time and the time for which the forecast is being made increases. The use of ensembles and model consensus help narrow the error and pick the most likely outcome. The increasing availability of atmospheric data during the last decades makes it important to find an effective and accurate tool to analyze and extract hidden knowledge from the huge data. Meteorological data mining is a form of data mining concerned with finding hidden patterns inside largely available meteorological data, so that the information retrieved can be transformed into usable knowledge [5]. Useful knowledge can play important role in understanding the climate variability and climate prediction. This motivated the authors to adopt data mining techniques in the nowcasting of snow/no-snow [26]-[35]. In the present research, the presence of snow/no-snow prediction is implemented with the use of empirical statistical technique, decision trees [35]. A decision tree is a classification scheme which generates a tree and a set of rules, representing the model of different classes, from a given dataset. The set of records available for developing classification methods is generally divided into two disjoint subsets-a training set and a test set. The former is used for deriving the classifier, while the latter is used to measure the accuracy of the classifier [35], [36] [38][44]. In this paper, we are introducing a new algorithm entitled Improved Snow Prediction Model (ISPM), an improvement to our earlier algorithms Snow Prediction Model (SPM), Improved Supervised Learning in Quest (ISLIQ), Supervised Learning using Gain Ratio as Attribute Selection Measure (SLGAS) and Supervised Learning using Entropy as Attribute Selection Measure (SLEAS) [38], [39], [41], [45]. We have laid out the rest of this paper as follows: Section II provides the related works. Section III introduces ISPM decision tree algorithm. Section IV provides the results obtained on twenty one international

locations using the ISPM, and these results have been compared with other decision tree models and finally Section V concludes the paper, followed by acknowledgements, references and authors biography.

II.

Related Works

Irene Y. H. Gu et al. [6], put forward a full automatic image analysis system for detection and analysis of snow/ice coverage on electric insulators of power lines using images which were captured by visual cameras in a remote outdoor laboratory test bed. Jinmei Pan et al. [7], put forth a passive microwave remote sensing techniques that detected wet snow in the south of china. Yajaira Mejia et al. [8], gave an approach for estimating the snowfall using neural networks on multi source remote sensing observations and ground based meteorological measurements. Melanie Wetzel et al. [9], projected a technique that supports the snowfall forecast and for the verification of radar limited mountainous terrain that includes matching the output parameters and graphics from high resolution mesoscale models to surface mesonets. Pascal Sirguey et al. [10], made use of ASTER and MODIS sensors, both on the TERRA platform by implementing the ARSIS concept so as to fuse the high spatial content of the two 250m spectral bands of MODIS into five 500m bands using wavelet based multi resolution analysis in the mountainous environment. Michael A. Rotondi [11] illustrated a Markov chain models across eight national weather stations using historical data from the global historical climatology network to predict a ‘snow day’. Gail M. Skofronick Jackson et al. [12], in their research interpreted how instruments like the W-band radar of Cloudsat, Global Precipitation Measurement Dual-Frequency Precipitation Radar ku- and Ka-bands, and the Microwave Imager can be used in the simulations of lake effect and synoptic snow events in order to determine the minimum amount of snow. Gail M. Skofronick Jackson et al. [13], demonstrated thresholds for detecting falling snow from satellite-borne active and passive sensors. Andrea Spisni et al. [14], presented an operational chain developed in the Emilia-Romagna region to monitor snow cover and snow water equivalent over the area managed by the Regional Catchment Technical Service. Alberto Martinez Vazquez et al. [15] presented an algorithm using GB-SAR imagery for the automatic recognition and classification of snow avalanches. Jeremie Bossu et al. [16], made use of a structure, based on computer vision which detects the presence of snow or rain. Noel Dacruz Evora et al. [17], used brightness temperature data, provided by seven channels SSM/I aboard the Defense Meteorological Satellite Program F11 and F-13 spacecrafts. Using which a modelling framework was put forth by combining passive microwave data, neural network based models and geostatistics for snow water equivalent retrieval and mapping.


International Review on Computers and Software, Vol. 10, N. 6

537


Hossein Zeinivand et al. [18], enforced a spatially distributed physically based model to detect snow and melting in the Latyan dam watershed in Iran. Xiaolan Xu et al. [19], developed a model that can be used for both active and passive microwave remote sensing of snow. B.B Fitzharris et al. [20], presented three case studies on the usage of satellite imagery for mapping seasonal snow cover in New Zealand, and also explored the effectiveness of using AVHRR imagery in order to obtain the presence of snow, snow covered area and snow line elevation on the mountain ranges of New Zealand. Ashok N. Srivastava et al. [21], in their research discussed the results based on kernel methods for unsupervised discovery of snow, ice, clouds and other geophysical processes based on data from the MODIS instrument. G. Singh et al. [22], developed a Radar Snow Index model to identify snow using SAR polarimetry techniques. In their research, full polarimetric L-band ALSOS-PALSAR data of snow cover area in Himalayan region have been analyzed based on various component scattering mechanism models and all model results are compared. Fan Ke et al. [23], developed a model to identify winter time heavy snow over Northeast China by using a inter annual increment prediction approach. Folorunsho Olaiya [24] investigated the use of artificial neural networks and decision tree algorithms in forecasting maximum temperature, rainfall, evaporation and wind speed using meteorological data collected from the city of Ibadan, Nigeria through Nigerian Meteorological Agency, Oyo state office. Manjeet Singh et al. [25] forwarded an attempt to develop an automatic technique for avalanche area identification and also its severity index. For the detailed relevant work refer our earlier papers [38], [39], [41], [45].

2. Sort T in ascending order and choose the initial attribute along with the associated class label. 3. Evaluate the interval range, as shown in Eq. (1):

Interval Range 

a max  a min Group size

(1)

where a max is the maximum value for the particular attribute, a min is the minimum value for the particular attribute and Group size is to be fixed by user. Upon the experimentation 2 is giving as maximum accuracy levels when compared with other levels. Based on the interval range, evaluate the split points and it is shown in Eq. (2) [38]. a. Initially check for change in the class label. b. If there is a change in the class label, evaluate the split points and the midpoint of changed class labels is the split point. For instance, Let V be the initial record and Vi be the second record: such that take Mid Point (V, Vi) only when there is change in the class label, shown in formula (2):

Split Point  Midpo int V ,Vi 

(2)

4. Choose the split point 1 and apply entropy attribute selection measure and evaluate the entropy value and continue this for all the split points obtained for initial attribute and the procedure is as follows: a. Initially, consider attribute and also along with its associated class label and evaluate attribute entropy and it is shown in formula (3) [38]: N

Attribute Entropy 



M



i 1



 Pj  Pi log 2 Pi  j 1

(3)



where Pi is the probability of class entropy belonging to class i. Logarithm is base 2 because entropy is a measure of the expected encoding length measured in bits. b. Further, consider class label and evaluate class entropy and is as follows: Class entropy is a measure in the information theory, which characterizes the impurity of an arbitrary collection of examples. If the target attribute takes on M different values, then the class entropy relative to this M-wise classification is defined in formula (4) [38]:

III. ISPM Decision Tree Algorithm Classification is the task of learning a target function f that maps each attribute set x to one of the pre-defined class labels y. The input for the classification is the training dataset, whose class labels are already known. Classification analyzes the training dataset and constructs a model based on the class label, and aims to assign a class label to the future unlabelled records [5], [35]. A set of classification rules are generated by such a classification process, which can be used to classify future data and develop a better understanding of each class in the database. Some of the classification models are decision trees, neural networks, genetic algorithms, statistical models like linear/geometric discriminates [35]. In the present research we are introducing a novel decision tree algorithm entitled ISPM for the now casting of snow/no-snow.

M

Class Entropy  

 Pi log 2 Pi

(3)

i 1

where Pi is the probability of class entropy belonging to class i. Logarithm is base 2 because entropy is a measure of the expected encoding length measured in bits. 5. Now, compute the entropy: it is the difference of class entropy and attribute entropy and is shown in formula (5) [38]:

III.1. Procedure for Evaluating the Split Points and Decision Tree Generation 1. Read the training dataset T.



538


Entropy  Cass Entropy  Attribute Entropy

The comparison in terms of split points is presented in Table II. Apparently, almost all results for ISPM are better than those of SPM. The results clearly show that the proposed ISPM reduces nearly 51.48 % when compared with SPM decision tree algorithm. Figure 1 presents the number of split points of the SPM and proposed ISPM algorithms graphically.

(5)

6. Once the entropy values are evaluated for all split points, choose the maximum entropy value as the best split point and continue this for the remaining attributes also. 7. Finally, if the number of attributes are N, we will get N best split points for individual attributes. As decision tree is a binary tree, there will be only one root node and for this reason, among the N entropy values choose one best entropy value to form the root node and it will be as follows: Consider all the attribute best split points along with entropy values. Choose the maximum entropy value is the best entropy value. Now, consider the maximum entropy value attribute as the root node and take its split point and divide the tree in binary format i.e. keep the values which are lesser to split point at the left side of the tree and keep the values which are greater and equals to the right side of the tree, and continue the process till it ends with a unique class label.

IV.

TABLE II SPLIT POINTS COMPARISON OF ISPM VS SPM City Name SPM ISPM Aberden 740 210 Bangkok 112 50 Barcelona 174 41 Benton 449 208 Botswana 195 88 Brazil 460 299 Cairo 165 162 Chennai 130 64 Delhi 281 162 Eglinton 360 43 Humberside 171 34 Hyderabad 116 72 Iceland 385 207 Lahore 190 51 Manchester 499 211 Norway 765 560 Olympia 449 349 Perth 246 136 Sellaness 391 154 Tirupathi 154 108 Valley 706 200

Results and Discussions

The proposed ISPM has been tested on 21 international locations historical datasets of snow/nosnow, collected from various meteorological departments [36]. We conducted experiments by implementing our proposed algorithm in Java Net Beans IDE 7.2. All experiments were performed on intel i3 core processor and 4 GB RAM with windows 7 operating system. We also divided our data set in to two parts: training set (75%), which is used to create the model, and a test set (25%), which is used to verify that the model is accurate and not over fitted [35]. Table I summarizes the characteristics of the datasets, arranged in alphabetical order, presenting the number of instances, training instances, testing instances, and classes. TABLE I SNOW/NO-SNOW DATASETS City Name Instances Training Aberdeen 6333 4750 Bangkok 5740 4305 Barcelona 6013 4510 Benton 23042 17281 Botswana 6047 4535 Brazil 6367 4775 Cairo 6143 4607 Chennai 6033 4525 Delhi 6015 4511 Eglinton 6318 4738 Humberside 1036 777 Hyderabad 5849 4387 Iceland 3512 2634 Lahore 4887 3665 Manchester 6338 4753 Norway 6105 4579 Olympia 23042 17281 Perth 6182 4636 Sellaness 5412 4059 Tiruptahi 6039 4529 Valley 6082 4561

Testing 1583 1435 1504 5761 1512 1592 1536 1508 1504 1580 259 1462 878 1222 1585 1526 5761 1546 1353 1510 1521

Fig. 1. Split points comparison of ISPM vs SPM

A common but poorly motivated way of evaluating results of Machine Learning Experiments is using specificity, accuracy, precision, dice and error rate. Specificity relates to the test's ability to exclude a condition correctly [35], [37]. Precision is defined as the proportion of the true positives against all the positive results. Dice is defined as proportion of two time true positives and the combination of true positives, false positives and false negatives [35], [37]. Prediction error is a measure of the performance of a model to predict the correct output, given future observations used as predictors. In order to reveal the performance of our proposed ISPM algorithm, we present comparison between SLIQ, SPM, SLGAS, ISLIQ and ISPM in terms of classification accuracy, sensitivity, specificity, precision, dice and error rate, using 21



539


international locations datasets collected from various meteorological departments. The comparison in terms of classification accuracy is presented in Table III. The proposed method yielded an average accuracy of 88.56%, better, when compared with SLIQ 87.74%, SPM 87.95%, SLGAS 88.28% and ISLIQ 88.34%. For some of the cities, the accuracy levels are more for other algorithms when compared with ISPM.

City Name Aberden Bangkok Barcelona Benton Botswana Brazil Cairo Chennai Delhi Eglinton Humberside Hyderabad Iceland Lahore Manchester Norway Olympia Perth Sellaness Tirupathi Valley

TABLE III ACCURACY COMPARISON OF ISPM WITH OTHER DECISION TREE MODELS SLIQ SPM SLGAS ISLIQ 87.3 85.47 85.97 87.61 96.09 94.49 95.19 98.11 95.8 95.14 95.67 96.07 70.05 70.14 72.03 70.12 93.78 96.16 93.58 95.43 75.5 73.05 75.75 73.36 88.99 89.7 89.77 89.98 76.65 76.12 77.51 76.35 96.14 94.94 96.8 93.15 89.24 90.06 90.06 89.56 93.05 94.59 94.98 93.82 96.5 97.8 94.79 96.4 89.17 88.49 90.2 88.49 84.82 86.05 85.89 84.65 92.74 92.87 89.58 93.43 88.99 90.89 90.62 90.89 70.05 70.14 72.03 70.12 94.3 94.43 96.31 94.43 75.9 77.67 79.45 84.4 97.54 97.41 97.41 97.48 90 91.38 90.32 91.3

algorithms graphically. As can be observed, ISPM obtained better results than SLIQ, SPM, SLGAS and ISLIQ i.e. larger specificity.

City Name Aberdeen Bangkok Barcelona Benton Botswana Brazil Cairo Chennai Delhi Eglinton Humberside Hyderabad Iceland Lahore Manchester Norway Olympia Perth Sellaness Tiruptahi Valley

ISPM 87.68 98.32 96.07 70.24 96.29 75.18 89.32 74.60 96.34 89.75 94.20 97.80 88.95 86.38 91.29 90.69 70.24 94.24 84.18 97.54 90.52

TABLE IV SPECIFICITY COMPARISON OF ISPM WITH OTHER DECISION TREE MODELS SLIQ SPM SLGAS ISLIQ 93.24 91.22 92.19 93.69 96.36 94.75 95.45 98.53 97.59 97.80 97.87 98.83 71.82 72.96 75.08 73.07 94.68 97.24 94.41 96.56 79.52 77.08 80.70 77.53 95.53 96.95 96.74 96.43 82.36 81.03 82.36 75.27 97.83 96.34 98.44 94.58 97.52 97.87 97.66 98.44 95.49 98.36 98.36 97.13 97.77 99.16 95.97 98.16 96.51 93.29 95.48 93.67 86.93 89.50 90.59 89.70 95.69 95.90 92.94 96.90 95.92 98.14 97.78 97.92 71.82 72.96 75.08 73.07 96.73 96.87 98.80 96.87 78.74 82.96 85.45 93.02 98.06 97.86 98.06 97.93 93.86 95.31 93.72 93.91

ISPM 93.24 98.74 98.83 72.28 97.30 79.00 96.10 78.74 97.83 98.72 97.13 99.16 96.00 89.50 94.01 97.71 72.28 96.73 92.42 98.06 94.00

But, on an average the ISPM model outperforms when compared with other algorithms, graphically shown in Fig. 2.

Fig. 3. Specificity comparison of ISPM with other decision tree models

The comparison in terms of precision is presented in Table V. Apparently, almost all precision results for ISPM are better than those of SLIQ, SPM, SLGAS and ISLIQ. For some of the cities, the precision levels are more for other algorithms when compared with ISPM. But, on an average the ISPM model outperforms when compared with other algorithms. The proposed method yielded an average precision of 31.41%, better, when compared with SLIQ 28.85%, SPM 29.46%, SLGAS 31.01% and ISLIQ 30.68%. Figure 4 presents the classification precision of the SLIQ, SPM, SLGAS, ISLIQ and proposed ISPM algorithms graphically. The comparison in terms of dice is presented in Table VI. Apparently, almost all dice results for ISPM are better than those of SLIQ, SPM, SLGAS and ISLIQ. For some of the cities, the dice are more for other algorithms when compared with ISPM. The proposed method yielded an average dice of 38.91%, better, when compared with SLIQ 38.45%, SPM 37.25%, SLGAS 38.45% and ISLIQ 37.24%.

Fig. 2. Accuracy comparison of ISPM with other decision tree models

The comparison in terms of specificity is presented in Table IV. For some of the cities, the specificity levels are more for other algorithms when compared with ISPM. But, on an average the ISPM model outperforms when compared with other algorithms. The proposed method yielded an average specificity of 92.27%, better, when compared with SLIQ 91.14%, SPM 91.59%, SLGAS 92.05% and ISLIQ 91.96%. Fig. 3 presents the classification specificity of the SLIQ, SPM, SLGAS, ISLIQ and proposed ISPM



540


TABLE V PRECISION COMPARISON OF ISPM WITH OTHER DECISION TREE MODELS City Name SLIQ SPM SLGAS ISLIQ ISPM Aberdeen 60.69 53.93 55.93 62.16 61.70 Bangkok 3.70 2.59 2.98 0.00 00.00 Barcelona 35.18 15.78 29.54 22.72 22.72 Benton 52.38 52.59 55.15 52.57 52.65 Botswana 12.22 18.00 12.63 13.55 20.00 Brazil 31.26 27.90 30.40 28.13 31.06 Cairo 22.22 17.30 22.03 22.12 21.42 Chennai 14.33 16.28 18.15 14.32 16.76 Delhi 3.03 8.47 8.00 4.76 11.11 Eglinton 46.96 56.52 56.00 51.11 55.00 Humberside 42.10 55.55 60.00 46.15 50.00 Hyderabad 3.03 0.00 3.33 14.12 14.81 Iceland 56.45 50.94 59.77 51.00 54.41 Lahore 54.16 57.76 58.14 54.78 58.43 Manchester 41.81 42.45 26.05 45.88 35.03 Norway 20.83 33.33 31.11 35.55 33.33 Olympia 52.38 52.59 55.15 52.57 52.65 Perth 9.25 9.61 21.73 9.61 07.54 Sellaness 31.19 30.52 32.66 42.95 42.48 Tiruptahi 6.45 8.57 0.00 8.82 06.45 Valley 6.31 8.10 12.50 11.42 12.12

error rate results for ISPM are better than those of SLIQ, SPM, SLGAS and ISLIQ. The proposed method yielded an average error rate of 11.44%, better, when compared with SLIQ 12.26%, SPM 12.05%, SLGAS 11.72% and ISLIQ 11.66%.

Fig. 5. Dice comparison of ISPM with other decision tree models

Fig. 6 presents the error rate of the SLIQ, SPM, SLGAS, ISLIQ and proposed ISLIQ algorithms graphically. TABLE VII ERROR RATE COMPARISON OF ISPM WITH OTHER DECISION TREE MODELS City Name Aberdeen Bangkok Barcelona Benton Botswana Brazil Cairo Chennai Delhi Eglinton Humberside Hyderabad Iceland Lahore Manchester Norway Olympia Perth Sellaness Tiruptahi Valley

Fig. 4. Precision comparison of ISPM with other decision tree models TABLE VI DICE COMPARISON OF ISPM WITH OTHER DECISION TREE MODELS City Name SLIQ SPM SLGAS ISLIQ ISPM Aberdeen 81.76 74.65 74.57 82.63 85.29 Bangkok 6.89 4.93 5.63 0.00 0 Barcelona 46.34 15.18 33.33 15.62 15.62 Benton 82.73 81.25 85.50 81.02 82.76 Botswana 20.95 26.86 22.01 20.77 30.3 Brazil 48.33 43.71 45.60 43.83 48.94 Cairo 19.25 10.77 15.29 15.63 16.75 Chennai 20.40 24.39 27.04 22.95 26.3 Delhi 3.38 12.34 8.00 7.47 13.55 Eglinton 30.84 39.79 42.21 24.46 23.91 Humberside 61.53 52.63 63.15 54.54 63.63 Hyderabad 3.84 0.00 5.12 24.11 25 Iceland 53.84 69.67 75.36 67.10 55.22 Lahore 91.49 92.06 86.84 80.51 94.6 Manchester 57.14 56.96 36.63 54.54 51.61 Norway 16.39 17.10 17.83 20.64 20.25 Olympia 82.73 81.25 85.50 81.02 82.76 Perth 10.75 10.98 16.12 10.98 8.6 Sellaness 51.14 44.73 45.55 44.85 46.59 Tiruptahi 10.25 14.28 0.00 14.63 10.25 Valley 7.59 8.75 16.25 14.91 15.38

Fig. 5 presents the classification dice of the SLIQ, SPM, SLGAS, ISLIQ and proposed ISPM tree algorithms graphically. The comparison in terms of error rate is presented in Table VII. Apparently, almost all

SLIQ 12.70 3.91 4.20 29.95 6.22 24.50 11.01 23.35 3.86 10.76 6.95 3.50 10.83 15.18 7.26 11.01 29.95 5.70 24.10 2.46 10.00

SPM 14.53 5.51 4.86 29.86 3.84 26.95 10.30 23.88 5.06 9.94 5.41 2.20 11.51 13.95 7.13 9.11 29.86 5.57 22.33 2.59 8.62

SLGAS 14.03 4.81 4.33 27.97 6.42 24.25 10.23 22.49 3.20 9.94 5.02 5.21 9.80 14.11 10.42 9.38 27.97 3.69 20.55 2.59 9.68

ISLIQ 12.39 1.89 3.93 29.88 4.57 26.64 10.02 23.65 6.85 10.44 6.18 3.60 11.51 15.35 6.57 9.11 29.88 5.57 15.60 2.52 8.70

ISPM 12.32 1.68 3.93 29.76 3.71 24.82 10.68 25.40 3.66 10.25 5.80 2.20 11.05 13.62 8.71 9.31 29.76 5.76 15.82 2.46 9.48

Fig. 6. Error rate comparison of ISPM with other decision tree models



541


As can be observed, ISPM obtained better results than SLIQ, SPM, SLGAS and ISLIQ i.e. lesser error rate.

V.

Point and so on that affect the presence of snow/no-snow can be identified using remote sensing of real-time satellite imagery.

Conclusion Acknowledgements

Nowcasting of snow/no-snow more accurately is the major research problem to most of the academicians, researchers, and scientists across the globe. In this paper, we presented a novel, efficient algorithm named ISPM decision tree algorithm, expanded to our previous algorithm SLEAS, SPM, SLGAS and ISLIQ for nowcasting the presence of snow/no-snow more accurately. Experimental results show that the ISPM algorithm scales up well to both large and small datasets with large number of attributes and class labels. We compare our proposed method with the SLIQ, SLEAS, SPM, SLGAS and ISLIQ decision tree algorithms in terms of the overall classification performance defined over five different performance measures namely accuracy, specificity, precision, dice and error rate. Results on the snow/no-snow 21 international locations datasets show that: a. the ISPM decision tree outperforms in terms of classification accuracy over 21 international locations of snow/no-snow datasets. The method yielded an average accuracy of 88.56%, better, when compared with SLIQ 87.74%, SPM 87.95%, SLGAS 88.28% and ISLIQ 88.34%. b. the ISPM decision tree outperforms in terms of classification specificity over 21 international locations of snow/no-snow datasets. The method yielded an average specificity of 92.27%, better, when compared with SLIQ 91.14%, SPM 91.59%, SLGAS 92.05% and ISLIQ 91.96%. c. the ISPM decision tree outperforms in terms of classification precision over 21 international locations of snow/no-snow datasets. The method yielded an average precision of 31.41%, better, when compared with SLIQ 28.85%, SPM 29.46%, SLGAS 31.01% and ISLIQ 30.68%. d. the ISPM decision tree outperforms in terms of classification dice over 21 international locations of snow/no-snow datasets. The method yielded an average dice of 38.91%, better, when compared with SLIQ 38.45%, SPM 37.25%, SLGAS 38.45% and ISLIQ 37.24%. e. the ISPM decision tree outperforms in terms of classification error rate over 21 international locations of snow/no-snow datasets. The method yielded an average error rate of 11.44%, better, when compared with SLIQ 12.26%, SPM 12.05%, SLGAS 11.72% and ISLIQ 11.66%. Future directions in this research include developing a decision tree algorithm for the nowcasting of snow/nosnow more accurately by reducing the computational complexity while evaluating the split points to a greater extent. In future, the most influencing parameters like Humidity, temperature, Pressure, Wind-Speed, Dew-

The authors would like to thank the referees for their valuable comments and suggestions, which greatly enhanced the clarity of this paper. We thank www.wunderground.com, for providing us the historical datasets of snow/no-snow. We are indebted to L V Narasimha Prasad for help and advice on the experimental analysis and paper setup. We are also grateful to the Management and Department of Computer Science and Engineering of Stanley College of Engineering & Technology for Women, Hyderabad and K L University, Guntur for providing their maximum support during the experimentation.

References [1] [2]

[3] [4] [5] [6]

[7]

[8]

[9]

[10]

[11]

[12]

[13]

[14]


Robert A. Houze: Cloud Dynamics. Academic Press (1994). Nabilah Filzah Mohd Radzuan, Andi Putra, Zalinda Othman, Azuraliza Abu Bakar and Abdul Razak Hamdan: Comparative Study – Three Artificial Intelligence Techniques for Rain Domain in Precipitation Forecast. International Journal of Environmental, Ecological, Geological and Mining Engineering (2013) 898-903. http://en.wikipedia.org/wiki/Weather_forecasting. http://en.wikipedia.org/wiki/Quantitative_precipitation_forecast. J. Han: Data Mining: Concepts and Techniques. Morgan Kaufmann Publishers. (2001) Irene Y.H. Gu, Unai Sistiag and Anders Fahlstrom: Online Detection of Snow Coverage and Swing Angles of Electrical Insulators on Power Transmission Lines. IEEE International Conference on Image Processing (2009) 3249 – 3252. Jinmei Pan, Lingmei Jiang and Lixin Zhang: Wet Snow Detection in the South of China by Passive Microwave Remote Sensing. IEEE International Geo Science and Remote Sensing Symposium (2012) 4863-4866. Yajaira Mejia, Hosni Ghedira and Reza Khanbilvardi: A Neural Network Based Approach for Multi-Spectral Snowfall Detection and Estimation, IEEE International Geo Science and Remote Sensing Symposium (2007) 2276 – 2279. Melanie Wetzel, Michael Meyers, Randolph Borys, Ray Mcanelly, William Cotton, Andrew Rossi, Paul Frisbie, David Nadler, Douglas Lowenthal, Stephen Cohn, and William Brown: Mesoscale Snowfall Prediction And Verification In Mountainous Terrain. AMS Journal Of Weather And Forecasting (2004) 806828. Pascal Sirguey, Renaud Mathieu Yves Arnaud, Muhammad M. Khan and Jocelyn Chanussot: Improved Resolution For The Detection Of Snow With Modis Using Wavelet Fusion. IEEE International Geo Science and Remote Sensing Symposium (2007) 3975-3978. Michael A. Rotondi: Estimating Transition Matrices to Predict Tomorrow’s Snowfall Using Real Data. Journal of Statistics Education (2010) 1-14. Gail M. Sko Fronick Jackson, Benjamin T. Johnson, and S. Joseph Munchak: Detection Thresholds of Falling Snow From Satellite-Borne Active And Passive Sensors, IEEE Transactions On Geo Science and Remote Sensing (2013) 4177-4189. Gail M. Skofronick Jackson and Benjamin T. Johnson: Thresholds Of Detection For Falling Snow From Satellite-Borne Active And Passive Sensors, IEEE International Geo Science And Remote Sensing Symposium (2007) 2637-2640. Andrea Spisni, Fausto Tomei, Sara Pignone, Enrico Muzzi, Alessandro Panzacchi, Gabriele Antolini, Giulia Villani, Michele Di Lorenzo, Rosanna Foraci1, Marco Bittelli and Erin S. Brooks:


542


[15]

[16]

[17]

[18]

[19]

[20]

[21]

[22]

[23]

[24]

[25]

[26] [27]

[28]

[29]

[30]

[31]

[32]

[33] [34]

Snow Cover Analysis in Emilia-Romagna. European Journal of Remote Sensing (2011) 59-73. Alberto Martinez Vazquez and Joaquim Fortuny Guasch: Snow Avalanche Detection and Classification Algorithm for GB-SAR Imagery, IEEE International Geo Science and Remote Sensing Symposium (2007) 3740-3743. Jeremie Bossu, Nicolas Hautière and Jean Philippe Tarel: Rain or Snow Detection in Image Sequences through Use of a Histogram of Orientation of Streaks, International Journal of Computer Vision (2011) 348-367. Noel Dacruz Evora, Ddominique Tapsoba and Danielle De Seve: Combining Artificial Neural Network Models, Geo statistics, and Passive Microwave Data for Snow Water Equivalent Retrieval and Mapping. IEEE Transactions on Geo Science and Remote Sensing (2008) 1925-1939. Hossein Zeinivand and Florimond De SmISLIQ: Simulation of Snow Covers area by a Physical Based Model. World Academy of Science, Engineering and Technology (2009) 428-433. Xiaolan Xu, Ding Liang, Leung Tsang, Konstantinos M. Andreadis, Edward G. Josberger, Dennis P. Lettenmaier, Donald W. Cline and Simon H. Yueh: Active Remote Sensing of Snow Using NMM3D/DMRT and Comparison With Clpx Ii Airborne Data, IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing (2010) 689-697. B. B. Fitzharris and B. P. Mcalevey: Remote Sensing of Seasonal Snow Cover in the Mountains of New Zealand Using Satellite Imagery. Taylor and Francis Geocarto International (1999) 3544. Ashok N. Srivastava and Julienne Stroeve: Onboard Detection of Snow, Ice, Clouds and Other Geophysical Processes Using Kernel Methods. Proceedings of the ICML 2003 Workshop on Machine Learning Technologies for Autonomous Space Applications (2003) 1-5. G. Singh, Y. Yamaguchi, S. E. Park and G. Venkataraman: Identification of Snow Using SAR Polarimetry Techniques. International Archives of the Photogrammetry, Remote Sensing and Spatial Information Science (2010) 146-149. Fan Ke and Tian Baoqiang: Prediction of Wintertime Heavy Snow Activity in North East China. Springer Chinese Science Bulletin (2013) 1420-1426. Folorunsho Olaiya: Application of Data Mining Techniques in Weather Prediction and Climate Change Studies, International Journal of Information Engineering and Electronic Business (2012) 51-59. Manjeet Singh, V. D. Mishra, N. K. Hakur and Jyoti Dhar Sharma: Remote Sensing GIS Based Statistical Modelling for the Prediction of Natural Hazards. International Journal of Engineering Research and Technology, (2012) 1-7. J.R. Quinlan: Induction of Decision Trees. Journal of Machine Learning (1986) 81-106. B. Chandra and P. Paul Varghese: Fuzzy Sliq Decision Tree Algorithm: IEEE Transactions on Systems, Man and Cybernetics (2008) 1294-1301. Anuja Priyama, Abhijeeta, Rahul Guptaa, Anju Ratheeb and Saurabh Srivastavab: Comparative Analysis of Decision Tree Classification Algorithms, International Journal of Current Engineering and Technology (2013) 334-337. Masud Karim and Rashedur M. Rahman: Decision Tree and Naive Bayes Algorithm for Classification and Generation of Actionable Knowledge for Direct Marketing. Journal of Software Engineering and Applications (2013) 196-206. Pedro Domingos and Michael J. Pazzani: Beyond Independence: Conditions for the Optimality of the Simple Bayesian Classifier. International Conference on Machine Learning (1996) 105-112. Manish Mehta, Rakesh Agarwal and Jorma Rissanen: SLIQ: A Fast Scalable Classifier for Data Mining. International Conference on Extending Database Technology (1996) 18-32. Rodrigo Coelho Barros, Marcio Porto Basgalupp, Andre C.P.L.F. De Carvalho and Alex A. Freitas: A Survey of Evolutionary Algorithms for Decision Tree Induction, IEEE Transactions on Systems, Man and Cybernetics (2012) 291-312. J.R. Quinlan: C4.5: Programs for Machine Learning, Morgan Kaufmann Publishers, (1993). S. Safavian and D. Landgrebe: A Survey of Decision Tree

[35] [36] [37]

[38]

[39]

[40]

[41]

[42]

[43]

[44]

[45]

Classifier Methodology, IEEE Transactions on Systems, Man And Cybernetics (1991) 660 -674. Arun k pujari: Data Mining Techniques. Universities Press (2004). http://www.wunderground.com/. Powers and M. W. David: Evolution: From Precision, Recall and F-Measure to Roc, Informedness, Markedness and Correlation. Journal of Machine Learning Technologies (2011) 37-63. Kishor Kumar Reddy C, Vijaya Babu B, Rupa C H: SLEAS: Supervised Learning using Entropy as Attribute Selection Measure. International Journal of Engineering and Technology (2014) 2053-2060. Kishor Kumar Reddy C, Rupa C H and Vijaya Babu B: A Pragmatic Methodology to Predict the Presence of Snow/NoSnow using Supervised Learning Methodologies. International Journal of Applied Engineering Research (2014) 11381-11394. Chandra, B., Mazumdar, S., Vincent, A. and Parimi, N: Elegant Decision Tree Algorithm for Classification in Data Mining. Proceedings of the Third International Conference on Web Information Systems Engineering (2002) 160−169. Kishor Kumar Reddy C, Rupa C H and Vijaya Babu: SPM: A Fast and Scalable Model for Predicting Snow/No-Snow. World Applied Sciences Journal (2014) 1561-1570. Pramote Luenam, Supawadee Ingsriswang, Lily Ingsrisawang, Prasert Aungsuratana, and Warawut Khantiyanan: A Neuro-Fuzzy Approach for Daily Rainfall Prediction over the Central Region of Thailand. International Multi Conference of Engineers and Computer Scientists (2010). Lily Ingsrisawang, Supawadee Ingsriswang, Pramote Luenam, Premjai Trisaranuwatana, Song Klinpratoom, Prasert Aungsuratana, and Warawut Khantiyanan: Applications of Statistical Methods for Rainfall Prediction over the Eastern Thailand. International Multi Conference of Engineers and Computer Scientists (2010). Thair Nu Phyu: Survey of Classification Techniques in Data Mining. International Multi Conference of Engineers and Computer Scientists (2009). Kumar Reddy, C., Rupa, C., Vijaya Babu, B., SLGAS: Supervised Learning Using Gain Ratio as Attribute Selection Measure to Nowcast Snow/No-Snow, (2015) International Review on Computers and Software (IRECOS), 10 (2), pp. 120129.

Authors’ information 1

Assistant Professor, Stanley College of Engineering & Technology for Women, Hyderabad, India. 2

Professor, K L university, Guntur, India.

C. Kishor Kumar Reddy obtained his B.Tech in Information Technology from JNTU Anantapur in 2011, M.Tech in Computer Science and Engineering from JNTU Hyderabad in 2013 and currently pursuing Ph. D in Computer Science Engineering from K L University, Guntur. He has published 28 papers in International Conferences and Journals, indexed by Scopus and Dblp databases. He has been awarded with Interscience Scholastic Award in the International Conference on Computer Science and Information Technology in 2012. He is the member of IEEE, CSI, ISTE, IAENG, UACEE, IACSIT, ISRD, INAAR, and IAPA. He is the editorial board member of IJAEIM. Dr B. Vijaya Babu is presently working as Professor in CSE department of K L University, VADDESWARAM. Guntur(D.T), Andhra Pradesh. He has obtained his B.Tech., (ECE) degree from College of Engineering, JNTU Kakinada Campus in the year 1993 and M.Tech(CSE) degree from College of Engineering, JNTU Anantapur Campus in the



543


year 2004. He obtained his PhD (CSSE) degree from A U College of Engineering, Andhra University, VISAKHAPATNAM, Andhra Pradesh in the year 2012.He has teaching experience of about 20 years, in various private engineering colleges of Andhra Pradesh, in various positions. His research area is Knowledge Engineering/Data Mining and published about 25 research papers in various International/Scopus journals. He is the life member of Professional bodies like ISTE.



544


Bandwidth Allocation in Wireless Mesh Network Using Efficient Path Selection Scheme Karunya Rathan Abstract – Wireless Mesh Network (WMN) is a type of communication network comprises of radio nodes formed in a mesh topology. WMN is an essential network to provide internet access and wireless connections to remote areas and metropolitan cities. Being a part of Internet services the mesh network must support extended multimedia applications to all its mesh clients. For a network it is necessary to afford efficient Quality-of-Service (QoS). Searching a new route with maximum bandwidth is one of the major troubles to support QoS in mesh networks. Owing to interference among various paths the network bandwidth is a well-known bottleneck metric in wireless mesh networks. In this research, a novel path weight algorithm is proposed in order to capture the information of available bandwidth. Also, an efficient path routing protocol is presented which is based on the novel path weight algorithm can satisfy the consistency and loopfreeness of the network. The consistency property assures that each mesh client creates a proper data packet forwarding decision; therefore a packet does travel through the correct path. The experimental results show that the proposed path weight protocol provides high throughput paths. Copyright © 2015 Praise Worthy Prize S.r.l. - All rights reserved.

Keywords: Wireless Mesh Networks, Mesh Topology, QoS and Bandwidth

I.

Generally, the interference is classified into two types: one is inter-flow interference and another one is intraflow interference [1], [3]. In Fig. 1, the upper path from A to D (A-B-C-D) possesses maximum available bandwidth than the middle (A-F-G-D) and lower paths (A-I-J-K) which is calculated according the formula given in [2] and [5] and is discussed in the following section. The next maximum bandwidth available path is the lower path from A to D (A-I-J-D) and the middle path from A-D (A-F-G-D) has low bandwidth when compared to the upper and lower paths. On applying the usual distance vector protocol to the Fig. 1, the node A just advertises the information of upper path to its neighbor nodes (or nearby nodes). According to the regular distance vector routing, node A only advertises the information of upper path to its neighbor nodes, thus the source node S cannot receive the maximum bandwidth path from itself to destination D.

Introduction

A Wireless Mesh Network (WMN) comprises of thousands of mesh nodes that are connected through wireless links in order to cover the entire service area whereas some of the mesh nodes require wired connection to the internet [1]-[27]. Being a part of worldwide internet, WMN must support extended multimedia applications to its users. While considering the Quality-of-Service wireless mesh network can provide efficient QoS [2]. Searching a new path with maximum network bandwidth is a major problem for supporting QoS in WMNs. The bandwidth of the new available path is defined as the maximum allowable rate of a data flow which can push through it to saturate its path. [1]. Thus, a path can allow the traffic rate of new data flow which is no greater than the bandwidth of the path. So, the new traffic flow will not disrupt the available bandwidth that assures the existing data flows. This paper mainly focuses to discover maximum bandwidth route from each source to each destination. While discussing about the maximum bandwidth trouble, it is necessary to talk about the Bandwidth Constrained Routing Trouble (BCRT) Maximum bandwidth is a sub problem of BCRT that is the problem of discovering a path with at least an afforded amount of usable bandwidth. [4]. Due to the interference of wireless transmission links, it is very difficult to find a maximum bandwidth available path between each source to each destination in the wireless networks.

. Fig. 1. Proposed Network Topology


545

Karunya Rathan

Still the source identifies the middle and lower path to D that has larger bandwidth. When A receives data packet from the source S, then it forward the packets to B but not to F and I by means of conventional destination based hop-by-hop packet routing since the upper path only has the maximum available bandwidth value than other (middle and lower paths) two paths. That is to say, the network packets actually do not transmitted on the maximum bandwidth path from source S to destination D. In literal fact, a proper packet routing protocols must guarantee the following two requirements such as consistency and optimality requirements. Building a routing metric is a key for designing an efficient bandwidth routing protocol. This paper proposes an efficient path routing protocol based on the novel path weight algorithm can capture the thought of maximum available bandwidth. Also the construction of distance and routing table is illustrated. Then a packet forwarding scheme is developed. The proposed system is compared with existing path weight routing protocols in order to show the performance of proposed protocol.

II.

Interference aware Resource Usage (IRU) is an extension of ETT metric that is weighted with the count of interference links. Contention Aware Transmission Time (CATT) extends the IRU metric by conceiving the result of rate of raw data on the links and data packet size since it require multiple channel links. There are several QoS routing protocols are proposed which work with the information of available bandwidth of each path [1], [3], [5], [13]-[16]. Liu et al. [14] presented a new link metric that provides an available bandwidth path of the channel link. It is divided into number of interference channel links on that path link. Some of the path selection processes are given in [3], [16]-[19] assumes the bandwidth requirement of a path connection request is well known.

III. Preliminaries In wireless mesh network a current flow of a path on the each path link is denoted by where is the available bandwidth of the link. If a node creates a new connection requires only passing through the link , can send at most K bits amount of data to the next link by not affecting existing packet flows. The study in [22] depicted how to get, and the next discussion assumes is well-known. Here the link estimator considers the bit error rate of a link; therefore the available bandwidth of a link turns the expected bandwidth of that link [24]. The bandwidth of path is calculated from the given formula [22], [23].

Related Work

Several researches have been conducted to find the widest path in which the researchers presented the novel path weights. If the path has maximum weight then it is recognized to be the maximum bandwidth path of that network. R. Draves et al. [6, D. Couto et al. [7] presented Expected Transmission Count (ETX) which is the predicted number of a path used for data packet transmission and the ETX count is needed to transmit a packet over a particular path. This count is estimated by occasionally transmitting an assigned path link probe data packet. The ETX metric of a particular path link is defined as the sum of entire ETX metric links on that path. This is one of the most former path link metric formulated and other upcoming metrics are stretched from it. [8]. Padhye et al. [9] presented a Routing in Multi-Radio, Multi- Hop Wireless Mesh Networks in which Expected Transmission Time (ETT) is discussed in a detailed manner. The ETT is an enhanced version of ETX that also regards the result of raw data rate and size of the data packets on the paths due to the usage of multiple channel links. In this paper, single channel wireless mesh network is considered and an assumption is made on the rate of raw data packets which is same on all the network links and all data packets are equal in size. In this instance, the Expected Transmission Time is same as Expected Transmission Count. Some other metrics such as iAWARE [10], IRU [11] and CATT [12] are the extensions of Expected Transmission Time. Interference Aware (iAWARE) is one of the metric of ETT which sets based on the count of interference links and the existing traffic loads depend on the interference paths.

B  p   min qQ p Cq

(1)

where:

1

Cq 

1

 B l  lq

The principle behind the above formula is the packet transmissions on all links in a clique cannot coincide with each other but it will occur in a consecutive manner. 1 Hence,  is the time taken to transmit 1 Mega B l  lq bit of information on all links in the clique q. The Cq is the available path bandwidth over the clique q. The maximum available bandwidth of a path link is the bottleneck clique bandwidth.

IV.

Efficient Path Routing Protocol

An efficient path routing protocol is proposed which is based on a mechanism called distance-vector. This mechanism provides the essential and sufficient condition to determine whether a network path is not effective which needs to be advertised. Then this illustrates a novel path weight that meets the optimality requirement.



546

Karunya Rathan

The efficient path routing is accomplished by implementing constant packet forwarding and the routing table is updated by this new path weight.

if:

1  p1   2  p2  , 2  p1   2  p2  3  p1   3  p2  and 4  p1   4  p2 

A. Efficient Path Selection The proposed path selection approach differs from the conventional path selection approaches. The traditional distance-vector mechanism is slightly modified to obtain novel distance vector mechanism. In traditional approach a node is assumed to report its efficient path to all its neighboring nodes by the way each and every node can find out its own efficient path. Broadcasting all path information to destinations is not a feasible solution. A node only can advertise the maximum bandwidth path from its own perception and its neighbor nodes might not able to discover its maximum bandwidth path.

where Bw  p  is the whole path bandwidth, BF  p  is the first link bandwidth, BT  p  is the sub path bandwidth that composed of first two links and BH  p  is the sub path bandwidth that composed of first three links. Definition 3: (Non-dominated paths). Given three   paths p1 , p2 and p3 , if   p1     p2  , we address path p1 dominates the path p2 and p3 . If we cannot detect a path dominating p1 , we mention path p1 is a non-dominated path.

B. Computation of Path Weight In order to compute path weight of each path a novel isotonic path weight technique is introduced which is mainly used to effectively construct the routing tables and is briefly explained in the following section. It is necessary to meet the consistency and optimality requirements of a path to reduce communication overhead. For that we require to use isotonicity [20], [21] property which can meet consistency and optimality requirements and is the essential and sufficient condition of a path weight in order to develop an efficient path routing protocol. This process is done by computing Composite Unused Bandwidth (CUB) by calculating isotonic path weight, bandwidth of a path and the proposed isotonic path weight. Definition 1: For Left Isotonicity, the quadruplet  N , , ,  is left-isotonic if   x     y  implies

C. Packet Forwarding The data packets are forward in a hop-by-hop manner and a constant hop-by-hop mechanism for packet forwarding is explained in this section. In a conventional hop-by-hop mechanism, the data packets hold the destination address only. When a node receives the data packet, it reads the destination address. Then, it looks for the succeeding hop by its destination address only. In our proposed mechanism, a packet carries destination address with addition of routing field which defines the succeeding four hops the data packet require to transmit. When a node receives this data packet, then it identifies the next hops or the path from the information given in the routing field of the packet. If the packet traverses through the hops given in the routing field, it automatically updates the routing fields after reaching the next hop in a periodic manner. In the proposed packet forwarding, the intermediate node only can decide the fourth succeeding hop but not the next single hop as mentioned in the traditional DSDV routing. Based on the routing table, each intermediate node makes route decision for packet forwarding. The routing table is maintained in each node contains the information of the first few hops of the path and it also comprises a routing field in each data packet. Thus, the proposed mechanism has similar characteristics of a hop-by-hop routing mechanism and it is also known as distributed packet forwarding approach. All non-dominated routes (one by one) from each source to each destination are directly related to the characteristics of routing protocol such as advertisement and space complexities. Let us take An and N  An  as

  z  x     z  y  , for all x, y,z  N , N denotes a number of paths,  is the concatenation operation of paths,  is the function that is used to map a network path to a weight, and  is the order relation. In Fig. 1, given paths p1 , p2 and p3 from S to D, assumes that upper path p1 has better higher bandwidth that p2 and p3 by comparing their path weights. Now, the proposed left-isotonic path weight known as Composite Unused Bandwidth (CUB) is given as follows: Definition 2: For a path p, the composite unused   bandwidth of path p, is denoted by   p  , where   p  is 1  p  ,2  p  ,3  p  ,4  p   in which:

the average number of neighbor nodes for each node and maximum number of non-dominated routes in the network. Consider a single non-dominating path is existed between source and destination that is passing through the same first three paths. Therefore, the proposed mechanism uses a polynomial-time packet routing algorithm in order to calculate the maximum throughput of each path.

1  p   Bw  p  , 2  p   BH  p  , 3  p   BT  p  and:

  4  p   BF  p  ,   p1     p2 



547

Karunya Rathan

The network consistency is discussed in the previous sections assume that each network node possesses the accurate state information about their neighbor nodes and the route updation leads to inconsistency. This inconsistency is entirely independent on routing metric i.e., routing mechanism though it is due to the route update delay of broadcast. Hence, this inconsistency occurs in all distributed packet routing protocols.

from A to D that are then dominated by N D  p  . The denoted path is the path from source to destination that is one hop away from N D  p  and then the source node estimates the CUB of that path. E. Route Updating Process After each transmission the network updates its routing table. When the network receives a new packet flow or broadcast an available link, the local available path bandwidth of each network node will vary. Thus, the maximum bandwidth available path from each source to each destination will differ. If the network threshold is smaller than the variation of local available path bandwidth of any node, then the node will advertise this new information to its neighbor nodes. The maximum bandwidth path from source to destination may change after obtaining the new bandwidth information. Though the network nodes are static, the information states vary very often. In traditional DSDV routing mechanism, each and every routing table entry is marked with a sequence number that is usually initiated by the destination node. Thus, the nodes can frequently separate stale routes from the new routes. The routing table is periodically updated by the network since all the network nodes periodically send updates quickly when important route information is present in the mesh network. If there are two route entries present from source to destination nodes, the source node always chooses the entry which is fresher which possesses larger sequence number and which is to be maintained in the routing table. In case the two routes possess the same network sequence number, then the route comparison is employed to find out that all paths should be maintained. Owing to the end-to-end delay presented during route update process, it is feasible that the path information maintained in some network node is in unpredictable manner. For instance, the maximum bandwidth available path that is maintained in the routing table might not be the widest anymore and the routing loops may takes place as well. This type of situation is denoted as inconsistency because of the temporary route updates. Therefore, the data packets are believed to be transmitted on the computed maximum bandwidth available path while all routing tables are fixed.

D. Constructing Table A path weight scheme is mainly based on the isotonicity property that allows developing an efficient path routing protocol which can discover the maximum bandwidth available path from each node to its corresponding destination separately. It is clearly shows that whether a path is effective to be advertised among the network nodes in order to verify whether the link is a potential sub-path of a maximum bandwidth available path. If a node discovers a new nondominated route, then it advertises this novel information to all its neighbor nodes in the process of routing. A route packet is a network packet that carries the information of a path. Non-dominated path is denoted as N D  p  and there are several non-dominated paths are existed between source and destination. The source node advertises this information to all its neighbor nodes through a route packet. N F  p  , N S  p  and NT  p  are the first, second and third next hops on non-dominated path N D  p  from source respectively. All nodes in the network can know its first four hops information of a path from the route packet which comprises all information of network nodes. This information is essential for consistent packet routing and is discussed in the following section. In our proposed scheme, each network node maintains two tables namely, distance table and routing table (see Tables I, II). All non-dominated paths are set by the source node (S) and are advertised by its neighbor nodes at the distance table. The routing table maintains the entire non-dominated path which is discovered by the source itself. In the proposed network, when source node obtains an advertisement from A that represents a non-dominated path N D  p  that is from node A to destination D then source node eliminates all the locally recorded routes

TABLE I DISTANCE TABLE Source

Destination

10

,

3

S

D

A

Non-dominated paths

   p

Neighbor

60

19

NS  p 

B

C

D

,4

F

G

D

, 16

I

J

D

, 5, 10

NT

 p

3

,

31 48

10

NF  p

60

,

31

,

48

20 7

,

19


48 7


548

Karunya Rathan

   p

Destination 40

40

,

17

D

120 48 25

,

40 48 13

 p

NS  p 

NT

 p

NU

 p

,8

A

B

C

D

,8

A

F

G

D

,8

A

I

J

D

9

19

,

NF

40

,

13

77

V.

TABLE II ROUTING TABLE Non-dominated paths

,

,

8 3

16 3

Performance Evaluation

Network Simulator-2 [25] is used for simulation and the performance of our proposed protocol for discovering the maximum bandwidth available path is evaluated. The proposed composite unused bandwidth is compared with the existing path weights. A. Routing Metrics The ETX [7] is the earliest routing metric to discover the maximum bandwidth available path link  l  which is defined as ETX l 

1 , where ptl is the probability of ptl

(a) Average Performance Ratio

packet loss of the link l on the MAC layer. The ptl is measured by transmitting dedicated path link probe packets. Couto et al. [7] presented the details on how to estimate ptl . Here the single channel mesh network is considered and it will not be compared with the routing metrics which are developed for multi-channel condition like ETT [6]. Interference-aware Resource Usage (IRU) [11] is another metric taken into the consideration and is given as IRU l  ETX  | Nel | , where Nel comprises of the neighbor nodes whose propagation interfere with the propagation on the link l. Since, it is assumed that all packets possess same size and all path links possess the same data rate.

(b) Throughput Figs. 2. Simulation of 75-nodes in 1000m×1000m(Scenario 1)

B. Simulation Results For the simulation, a random network topology is generated in NS-2 simulator. The minimum hop count is measured which is the distance between any two nodes. In the random network, some possible node pairs are taken into consideration. The proposed CUB, IRU and ETX may discover different paths (those minimum hop counts are calculated) between the possible node pairs. The proposed protocol can also provide an assessment for the maximum available bandwidth of the widest path. Then a new flow on that path in order to measure the throughput and the new data flow possesses high data rate larger than the available path bandwidth of the widest path is established, therefore, the maximum throughput maintained by that path without disrupting the bandwidth supported for existing packet flows is obtained.

The throughput of the paths discovered by the proposed CUB, ETX and IRU protocols are compared in order to show the performance of each protocol to discover the maximum bandwidth path. To perform simulation 75 nodes are deployed in 1000m×1000m and there are 300 bidirectional links that exist in the random network in which only 75 links in a random manner are selected and the existing one-hop packet flow is established on them. The minimum hop count of each node pair is estimated and we 10 node pairs are chosen so that each node pair possesses same path distance. In this random network topology, it is considered that the path distance of each node pair 3 to 10 so there are totally 130 multihop packet flows exist.



549

Karunya Rathan

Fig. 2(a) shows the average performance ratio of existing and proposed metrics as a function of hopdistance (Scenario 1). From the figure it is observed that the proposed protocol has best finding path in terms of high throughput. Fig. 2(b) shows the simulation result of the packet flows of existing and proposed protocols. From the figures it is clearly observed that the existing protocols ETX and IRU do not provide better performance in some cases. Back ground traffic is another routing metric determines the network performance; and the back ground traffic is varied in order to measure the performance of existing and proposed protocols. Next the performance of the network is estimated by allowing the data rates of existing packet flows with 140 Kbps and the simulation result is shown in Fig. 3(a) (Scenario 2) and its corresponding throughput is shown in Fig. 3(b).

VI.

Conclusion

Thus, an efficient path routing protocol is presented which is mainly based on a novel path weight algorithm. The efficient path routing is performed by choosing an efficient path, calculating each path weight, building distance table and routing table, forwarding data packets with consistency and optimality requirements and finally by route updation. There by choosing a maximum bandwidth path from each source to each destination. By using hop-by-hop packet routing the overall network throughput is increased which utilizes the information of available path bandwidth. Thus, packet dropping ratio is reduced and packet delivery ratio is increased. The proposed protocol is compared with the existing path weight routing protocols and the simulation results shows that the proposed protocol provides better performance by means of reduced end-to-end delay, control overhead and increased packet delivery ratio.

References [1]

[2]

[3]

[4] (a) Average Performance Ratio [5]

[6]

[7]

[8]

[9]

[10] (b) Throughput Figs. 3. 75-nodes in 1000m×1000m with exiting packet flows following 140Kbps (Scenario 2)

[11]

By comparing the scenario 1 and scenario 2, the scenario 1 has reduced background traffic and increased or maximum available bandwidth that the scenario 2. Thus, our simulation results show that the proposed routing metric (CUB) provides better performance in discovering maximum throughput path than the existing routing metrics

[12]

[13]


T. Salonidis, M. Garetto, A. Saha, and E. Knightly, “Identifying High Throughput Paths in 802.11 Mesh Networks: A ModelBased Approach,” Proc. IEEE Int’l Conf. Network Protocols (ICNP ’07), pp. 21-30, Oct. 2007. Q. Zhang and Y.-Q. Zhang, “Cross-Layer Design for QoS Support in Multihop Wireless Networks,” Proc. IEEE, vol. 96, no. 1, pp. 234-244, Jan. 2008. J. Tang, G. Xue, and W. Zhang, “Interference-Aware Topology Control and QoS Routing in Multi-Channel Wireless Mesh Networks,” Proc. ACM MobiHoc, pp. 68-77, May 2005. C.-Y. Chiu, Y.-L. Kuo, E. Wu, and G.-H. Chen, “BandwidthConstrained Routing Problem in Wireless Ad Hoc Networks,” IEEE Trans. Parallel and Distributed Systems, vol. 19, no. 1, pp. 4-14, Jan. 2008. H. Li, Y. Cheng, C. Zhou, and W. Zhuang, “Minimizing End-toEnd Delay: A Novel Routing Metric for Multi-Radio Wireless Mesh Networks,” Proc. IEEE INFOCOM, pp. 46-53, Apr. 2009. R. Draves, J. Padhye, and B. Zill, “Comparison of Routing Metrics for Static Multi-Hop Wireless Networks,” Proc. ACM SIGCOMM, pp. 133-144, Sept. 2004. D. Couto, D. Aguayo, J. Bicket, and R. Morris, “A HighThroughput Path Metric for Multi-Hop Wireless Routing,” Proc. ACM MobiCom, pp. 134-146, Sept. 2003. M. Campista, D. Passos, P. Esposito, I. Moraes, C. Albuquerque, D. Saade, M. Rubinstein, L. Costa, and O. Duarte, “Routing Metrics and Protocols for Wireless Mesh Networks,” IEEE Network, vol. 22, no. 1, pp. 6-12, Jan. 2002. R. Draves, J. Padhye, and B. Zill, “Routing in Multi-Radio, Multi- Hop Wireless Mesh Networks,” Proc. ACM SIGCOMM, pp. 114- 128, Oct. 2004. A.P. Subramanian, M.M. Buddkihot, and S. Miller, “Interference Aware Routing in Multi-Radio Wireless Mesh Networks,” Proc. Second IEEE Workshop Wireless Mesh Networks (WiMesh ’06), pp. 55- 63, Sept. 2006. Y. Yang, J. Wang, and R. Kravets, “Designing Routing Metrics for Mesh Networks,” Proc. IEEE Workshop Wireless Mesh Networks (WiMesh ’05), Sept. 2005. M. Genetzakis and V.A. Siris, “A Contention-Aware Routing Metric for Multi-Rate Multi-Radio Mesh Networks,” Proc. Fifth Ann. IEEE Comm. Soc. Conf. Sensor, Mesh and Ad Hoc Comm. And Networks (SECON ’08), pp. 242-250, 2008. H. Li, Y. Cheng, and C. Zhou, “Multi-Hop Effective Bandwidth Based Routing in Multi-Radio Wireless Mesh Networks,” Proc. IEEE Global Telecomm. Conf. (GlobeCom ’08), pp. 1-5, Nov. 2008.


550

Karunya Rathan

[14] T. Liu and W. Liao, “Interference-Aware QoS Routing for MultiRate Multi-Radio Multi-Channel IEEE 802.11 Wireless Mesh Networks,” IEEE Trans. Wireless Networks, vol. 8, no. 1, pp. 166- 175, Jan. 2009. [15] L. Chen and W.B. Heinzelman, “QoS-Aware Routing Based on Bandwidth Estimation for Mobile Ad Hoc Networks,” IEEE J. Selected Areas in Comm., vol. 23, no. 3, pp. 561-572, Mar. 2005. [16] Q. Xue and A. Ganz, “Ad Hoc QoS On-Demand Routing (AQOR) in Mobile Ad Hoc Networks,” J. Parallel and Distributed Computing, vol. 63, pp. 154-165, 2003. [17] W. Liao, Y. Tseng, and K. Shih, “A TDMA-Based Bandwidth Reservation Protocol for QoS Routing in a Wireless Mobile Ad Hoc Networks,” Proc. IEEE Int’l Conf. Comm. (ICC ’02), pp. 3186-3190, Apr. 2002. [18] K. Shih, C. Chang, Y. Chen, and T. Chuang, “Dynamic Bandwidth Allocation for QoS Routing on TDMA-Based Mobile Ad Hoc Networks,” Computer Comm., vol. 29, pp. 1316-1329, 2006. [19] C. Zhu and M.S. Corson, “QoS Routing for Mobile Ad Hoc Networks,” Proc. IEEE INFOCOM, pp. 958-967, June 2002. [20] Y. Yang and J. Wang, “Design Guidelines for Routing Metrics in Multihop Wireless Networks,” Proc. IEEE INFOCOM, pp. 22882296, Apr. 2008. [21] J.L. Sobrinho, “Algebra and Algorithms for QoS Path Computation and Hop-by-Hop Routing in the Internet,” Proc. IEEE INFOCOM, pp. 727-735, Apr. 2001. [22] Y. Yang and R. Kravets, “Contention-Aware Admission Control for Ad Hoc Networks,” IEEE Trans. Mobile Computing, vol. 4, no. 4, pp. 363-377, Apr. 2009. [23] H. Li, Y. Cheng, C. Zhou, and W. Zhuang, “Minimizing End-toEnd Delay: A Novel Routing Metric for Multi-Radio Wireless Mesh Networks,” Proc. IEEE INFOCOM, pp. 46-53, Apr. 2009. [24] H. Zhai and Y. Fang, “Impact of Routing Metrics on Path Capacity in Multirate and Multihop Wireless Ad Hoc Networks,” Proc. 14th IEEE Int’l Conf. Network Protocols (ICNP ’06), pp. 86-95, Nov. 2006. [25] The Network Simulator—ns2, http://www.isi.edu/nsname/ns, 2011 [26] Kumar Giri, R., Saikia, M., Multipath routing for admission control and load balancing in wireless mesh networks, (2013) International Review on Computers and Software (IRECOS), 8 (3), pp. 779-785. [27] Snehalatha, N., Rodrigues, P., A Clique Based Bandwidth Computation for Consistent Routing in Wireless Mesh Networks, (2013) International Journal on Information Technology (IREIT), 1 (3), pp. 200-205.

Authors’ information Asst. Professor, Faculty of computing, Sathyabama University. Tel.: +91 9600663238 E-mail: [email protected] Mrs. Karunya Rathan has an overall teaching experience of seven years, of which she is working for the Sathya bama University for the past 5 years, where she finished her M.Tech.



551


A New Low Complexity Transform for Peak Power Reduction in Discrete Hartley Transform Based Multicarrier OFDM System N. Bharathi Raja1, N. Gangatharan2 Abstract – In this paper, a new low complexity transform which combines the fast Walsh Hadamard transform (WHT) and fast Hartley transform (FHT) into a single fast orthonormal transform is proposed for orthogonal frequency division multiplexing (OFDM) across additive white Gaussian noise (AWGN) channel model. The new transform is developed through the sparse matrices method in OFDM system, which is capable of reducing peak to average power ratio (PAPR) of the transmitted symbols with better bit error rate (BER) performance degradation at a reasonable reduced complexity. The system performance is verified via simulations. Compared with the FHT method, the proposed OFDM signal can be generated by FHT via WHT with lower PAPR and also the computational complexity nearly halved. It reveals that the proposed OFDM has the same BER performance as FHT, but proposed OFDM transceiver shows superiority on computational complexity. The proposed scheme is a cost-effective and efficient multicarrier modulation scheme. Copyright © 2015 Praise Worthy Prize S.r.l. - All rights reserved.

Keywords: OFDM, Hartley Transform, FHT & WHT

An alternative approach for mitigating the PAPR problem is based on signal transformation. Most of OFDM transceivers employ inverse fast Fourier transform (IFFT) and FFT to perform OFDM modulation and demodulation in transmitter and receiver, respectively. The FFT/IFFT has been one of the most critical modules in OFDM transceivers. The rapidly increasing demand of OFDM based applications for wireless broadband communications makes processing speed an important major consideration in FFT architecture design [9]. Even so, nowadays most FFT researches focus on the hardware implementation techniques instead of algorithm study since the FFT algorithms have already been well developed. However, IFFT/FFT is not the only orthogonal basis for OFDM systems. Recently, studies focus on the different trigonometric transforms, such as discrete Hartley transform (DHT) and discrete cosine/sine transform (DCT/DST) have been introduced to be an alternative orthogonal transformation for OFDM systems. The DHT based OFDM is attractive for its intimate relation to DFT. Hartley transform is particularly attractive for the processing of real signals [10]. Fourier transform always implies a complex processing and the phase carries fundamental information, while Hartley transform is a real trigonometric transform. In [11], a DHT based OFDM is proposed to transmit the signal over the subcarriers separately. In this paper, a new low complexity transform to combine the Walsh Hadamard transform (WHT) and the fast Hartley transform (FHT) into a single orthonormal unitary transform is proposed in OFDM system to achieve significant BER performance and considerable PAR reduction.

Nomenclature AWGN BER CCDF CP DCT/DST DHT FHT IFFT OFDM PAPR SNR WHT

Additive White Guassian Noise Bit Error Rate Complementary Cumulative Distribution Function Cyclic Prefix Discrete Cosine/Sine Transform Discrete Hartley Transform Fast Hartley Transform Inverse Fast Fourier Transform Orthogonal Frequency Division Multiplexing Peak To Average Power Ratio Signal To Noise Ratio Walsh Hartley Transform

I.

Introduction

Orthogonal Frequency Division Multiplexing (OFDM) is a multicarrier modulation (MCM) scheme in which multiple data streams are modulated with individual orthogonal subchannels [1]-[24]. Due to the advantage of spectral efficiency and the immunity of multipath channel, OFDM systems are widely adopted in many wireless communication standards [2]. However, the major drawback of OFDM is high peak to average power ratio (PAPR). Several methods have been developed to address the PAPR problem [3]-[8]. These different techniques provide reduced PAPR at the expense of increased system complexity, reduced spectral efficiency, and performance for improved linearity.


552

N. Bharathi Raja, N. Gangatharan

Input Data

Serial to Parrallel convertor

Symbol Mapping

Parrallel to serial convertor

IFHT via WHT

Add CP

Channel

Output Data

Demapping

Parrallel to serial convertor

FHT via WHT

Serial to Parrallel convertor

Remove CP

Fig. 1. Block diagram of proposed system model

Compared with FFT scheme, it can reduce computational complexity by half. The remainder of this paper is organized as follows. In section II, the proposed OFDM system model is presented and also discussed a low complexity combined WHT-DHT transform structure. The system performance over AWGN channel is simulated in section III. Finally, conclusions are presented in section IV.

II.

y  h xn

At the receiver section, the received signal vector

rn  ,n  0,N  1T

Hk 

1 N

R=Ty II.1.

Proposed System Model

1 N

(5)

DHT Via WHT Transform Structure

In this paper, the proposed transform, which is used in the receiver of Fig. 2 is evaluated as:

T

1 HW N

(6)

H is the normalized N  N DHT matrix rearranged by row reverse order, W is the Walsh-Hadamard matrix. Consequently, the H matrix in row reverse order can be computed as:

N 1

 H k cas  2 nk / N ,

is fed into the proposed transform

block to signal as:

The block diagram of OFDM system based on FHT via WHT is depicted in Fig. 1. The FHT via WHT scheme combining DHT and WHT is proposed in this work. The N point inverse FHT and FHT are defined by [12]-[13]:

xn 

(4)

n  0,1, ,N  1 (1)

k 0 N 1

 xn cas  2 nk / N ,

k  0,1, ,N  1 (2)

n 0

HN

where, cas    cos    sin   .

 AN   2  BN  2

AN  2    BN  2 

(7)

xn is the IDHT sequence and Hk is the DHT sequence. In proposed OFDM system, the modulating bits are modulated on BPSK, which are fed into the IDHT via WHT to compute the discrete time real baseband signal as:

where, A and B are sub matrices of H. On the other hand, the WHT of the product of two sequences is equivalent to the dyadic convolution of their WHT, it means [14]-[15]:

X n  T ' xn

W  X 1  X 2   WX 1  WX 2

(3)

(8)

where: X1 and X2 are discrete data vectors in the frequency domain and  represents dyadic convolution [16], [17]. In general, the W matrix can be written as a function of lower order matrices as:

where, W and H’ are WHT matrix and the normalized N  N IDHT matrix rearranged by column reverse order. 1 T '  WH ' is conversion matrix. In proposed OFDM N transmitter, N point IDHT via WHT is performed, instead of applying N point IDHT. The resulting symbols are appended with cyclic prefix (CP). Consequently, the received signal after removing the CP can be written as:

W N  WN   2 W N  2


WN  2    WN  2 

(9)


553


By substituting (9) and (7) in (6), we get:

1 TN  N

 2 AN W N  2 2  0 

0

   2BN WN  2 2 

(10)

To show the block diagonal structure of TN, the normalized N  N Hartley matrix can be written as:

HN

1 1 1   cas  2   1 cas 1 1   1 cas  2  cas  4   8       1 cas  N  1 cas  2  N  2   

  cas  N  1   cas  2  N  1      cas   N  1 N  1   1

(11)

For instance, the case N=8:

1  1 1  1 1  H8  8 1  1 1  1 

1

1

1

1

1

1

1

cas 1 cas  2 

cas  3

cas  4 

cas  5 

cas  6 

cas  2  cas  4 

cas  6 

cas  8 

cas 10 

cas 12 

cas  3 cas  6 

cas  9 

cas 12  cas 15 

cas 18 

cas  4  cas  8 

cas 12  cas 16  cas  20 

cas  24 

cas  5  cas 10  cas 15  cas  20  cas  25 

cas  30 

cas  6  cas 12  cas 18  cas  24  cas  30 

cas  36 

cas  7  cas 14 

cas  42 

cas  21 cas  28  cas  35 

   cas 14    cas  21   cas  28    cas  35   cas  42    cas  49   cas  7 

(12)

when rearranging the DHT matrix H 8 by row reversed order, can be obtained as:

1 1  1  1 1 H8  8 1  1 1  1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1.4142  1.4142

1

1

1 1 1  1 A4   1 1  1  1

1

0

1

 1.4142

1

0

1

1.4142

1

0

1

 1.4142

1

0

1

0

1

1.4142

1

0

1

1

Consequently, the above matrix can be written in terms of sub matrices as:

1  A4 H8   8  B4

1

A4   B4  1 1  1   1  1   1 1

1 1 B4   1  1

(14)

1

  1   1  1   0  0  1.4142    1.4142 

1.4142

1

 1.4142 0

1 1

0

1

(13)

0

  0   1.4142   1.4142 

(16)

Furthermore, the sub matrices in (15) and (16) can be subdivided into their internal submarines as:

1

(15)

A ' A4   2  A2 '


A' 2 '   A2 ' 

(17)


554


B ' B4   2  B2 ''

B2 ''   B2 ' 

(18)

1 A2 '   1

1  1

(19)

1 B2 ''   1

0 0 

(20)

1  1.4142  B2 '    1 1.4142 

The fast proposed transform requires  N log 2 N  2  N  1  of real multiplications. 2 Transform algorithms are compared based on the total number of arithmetic operations for 64 subcarriers as shown in Table I. TABLE I A COMPARISON OF REAL MULTIPLICATIONS FOR VARIOUS TRANSFORMS FOR 64 SUBCARRIERS Example Fast Transforms Real Multiplications (Real MUX) N point IFFT

(21)

N point IFHT

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

Proposed Transform

1  1  1   1 (22) 1   1 1   1 

W4   W4 

 2 B4W4 

TABLE II SIMULATION PARAMETERS FOR OFDM SYSTEMS Simulation Parameters Values FFT/FHT 64 CP 16 Modulation 16-QAM Channel AWGN

(24)

In order to verify the validity of our analytically derived technique, MATLAB simulation program is performed. The simulation is carried out for the following metrics:  Complementary Cumulative Distribution Function (CCDF);  BER performance. Fig. 2 shows a power spectral density for 64 subcarriers DHT and DFT based OFDM systems. To analyze the performance of the OFDM system, amplitude of DFT-OFDM signal and DHT-OFDM scheme is obtained and plotted in Fig. 3(a) and Fig. 3(b), respectively. The proposed scheme can enlarge small signals and change the signal peaks, which leads to a higher peak power level of output. This scheme provides reduction of PAPR by extra enlargement of amplitudes. Fig. 4 illustrates the CCDF plot of DFT-OFDM, DHTOFDM and proposed OFDM signal with 64 subcarriers.

T8  1 0  0  0 0  0 0  0

0 1 0 0 0 0 0 0

0 0 1 0 0 0 0 0

0 0 0 0 0 0 1 0 0 0.8536 0 0.1464 0 0.3536 0  0.3536

0 0 0 0 0.1464 0.8536  0.3536 0.3536

129

2

To study the performance of the modified T-OFDM, this section illustrates the performance analysis of the proposed OFDM, DHT-OFDM and the DFT-OFDM systems across AWGN channel model. Further, the simulation parameters required to investigate the performances of the proposed OFDM illustrated in Table II.

(23)

0

 N log 2 N  2  N  1

III. Simulation Results

To show the block diagonal structure of T8, the matrix can be written as (24)-(25):

1 2 A W T8   4 4 8 0

192

Clearly, the results show that the proposed transform involves less computational complexity than the direct computation of DHT.

Consequently, the above Eq. (22) can be written in terms of sub matrices as:

W W8   4 W4

384

2

The normalized 8×8 WHT matrix can be written as:

1 1 1 1 1  1 1 1  1 1 1 1  1  1 1 1 W8   1 1 1 1  1 1 1  1 1 1 1 1  1  1 1 1 

N log 2 N N log 2 N

0 0   0 0   0 0  0 0  0.3536  0.3536    0.3536 0.3536  0.1464 0.8536   0.8536 0.1464  (25)

Hence, the proposed transform based on the block diagonal structure reduces the superposition of the subchannels, as seen in (25).



555


model. It can be shown that the proposed OFDM improves the BER performance significantly than the conventional DFT-OFDM system. As can be seen from Fig. 5, the proposed OFDM system achieves SNR about 30dB at 10-3 BER level which has better BER performance than DHT-OFDM system. Hence, the proposed OFDM system outperforms the conventional OFDM in terms of PAPR reduction, significant complexity reduction and better BER performance.

25 DFT-OFDM DHT-OFDM 20

spectrum signal

15

10

5

0 0

10

-8

-6

-4

-2

0 frequency

2

4

6

8

DFT-OFDM DHT-OFDM Proposed OFDM

10

CCDF =Pr(PAPR>PAPR0)

-5 -10

Fig. 2. Spectrum representation of DFT OFDM and DHT OFDM DFT-OFDM 1.4

1.2

-1

10

Amplitude

1

0.8 -2

10

3.5

4

4.5

5

0.6

0.4

5.5 PAPRO (dB)

6

6.5

7

7.5

Fig. 4. CCDF plot for DFT OFDM, DHT OFDM and proposed OFDM signals with 64 subcarriers

0.2 0

10 0

0

1000

2000

3000 4000 5000 Subcarrier Index

6000

7000

DFT-OFDM DHT-OFDM Proposed OFDM

8000

-1

Fig. 3(a). Time domain representation of DFT OFDM signal

10

DHT-OFDM

BER

1 0.9

-2

10

0.8 -3

0.7

10

Amplitude

0.6 0.5 -4

10

0.4

0

5

10

0.3

15 SNR(dB)

20

25

30

Fig. 5. BER performance of DFT OFDM, DHT OFDM and proposed OFDM signals with 64 subcarriers

0.2 0.1 0

0

1000

2000

3000 4000 5000 Subcarrier Index

6000

7000

8000

IV.

Conclusion

In this paper, we have proposed a cost effective and efficient low complexity transform based OFDM system. Compared with FHT scheme, the fast FHT via WHT can generate the OFDM signal with lower PAPR and reduced computational complexity. Based on the sparse block diagonal transform structure, the proposed OFDM system can achieve PAPR reduction by a range of 0.8dB

Fig. 3(b). Time domain representation of DHT OFDM signal

In Fig. 4, it can be seen that the PAPR of proposed OFDM achieves 6dB which is less than that of conventional DHT- OFDM system by range of 0.8 dB. Fig. 5 illustrates the BER performance of DFT- OFDM, DHT-OFDM and proposed OFDM over AWGN channel Copyright © 2015 Praise Worthy Prize S.r.l. - All rights reserved


556


than that of DHT OFDM system. The simulation results also indicate that the proposed OFDM system has a better BER performance than the DHT-OFDM system. Furthermore, the size of the DHT via WHT scheme was demonstrated to be the same as the number of subchannels, thus there is no data rate losses when utilizing the T-transform with the multicarrier transmission techniques. Simulation results were shown that the proposed OFDM system outperforms the OFDM system in the presence of AWGN channel, and also minimize considerable PAPR. Consequently, a combined DHT via WHT OFDM system will benefit from the reduced PAPR, SNR and no BER performance degradation. For N = 64, the proposed OFDM achieves 1.4dB PAPR reduction over the conventional DFT-OFDM. Consequently, the proposed transform based OFDM schemes will benefit from the improved BER performance, reduced PAPR level and computational complexity reduction.

[15]

[16]

[17]

[18]

[19]

[20]

[21]

References [1]

[2]

[3]

[4]

[5]

[6] [7]

[8]

[9]

[10]

[11]

[12]

[13]

[14]

[22]

H. Liu and G. Li, OFDM-based broadband wireless networks; Design and optimization. New Jersey, USA: John Wiley and Sons, Inc., 2005. Deepa, T., Kumar, R., Low complexity, high throughput layered FFT structure for BI based COFDM systems, (2013) International Review on Computers and Software (IRECOS), 8 (5), pp. 1180-1185. M. Park, H. Jun, J. Cho, N. Cho, D.Hong, and C. Kang, “PAPR reduction in OFDM transmission using Hadamard transform,” in Proc. IEEE ICC, New Orleans, LA, USA, pp. 430–433, June 2000. Han, S.H., et al.: ‘An overview of peak-to-average power ratio reduction techniques for multicarrier transmission’, IEEE Wirel. Commun., 12, (2), pp. 56–65 ,2005. T. Jiang and Y. Wu, “An overview: peak-to-average power ratio reduction techniques for OFDM signals,” IEEE Trans. Broadcast., vol. 54, no. 2, pp.257–268, June 2008. Baig, I., et al.: ‘PAPR analysis of DHT-precoded OFDM system for M-QAM’. Proc. 2010 ICIAS, Malaysia, June 2010. Muquet, Z. Wang, G. B. Giannakis, M. de Courville, and P. Duhamel, “Cyclic prefixing or zero padding for wireless multicarrier transmissions,” IEEE Trans. Commun., vol. 50, no. 12, pp. 2136–2148, Dec. 2002. S. Weinstein and P. Ebert, “Data Transmission by FrequencyDivision Multiplexing Using the Discrete Fourier Transform,” IEEE Trans. Commun. Technol., vol. 19, no. 5, pp. 628–634, Oct. 1971. Arioua, M., Belkouch, S., M'rabet Hassani, M., Efficient 64-point FFT/IFFT processor based on 8-point FFT module for OFDM based WLAN, (2012) International Review on Computers and Software (IRECOS), 7 (1), pp. 92-99. Sorensen, H.V, Houston, Jones D.L, “ On Computing the Discrete Hartley Transform”, IEEE Trans. Acoustics, Speech and Signal Processing,Vol.33,no.5,pp.1231-1238,Oct 1985. C. Wang, et al., “Discrete Hartley transform based multicarrier modulation,” in Proc. IEEE ICASSP, Istanbul, pp. 2513–2516, 2000. M. S. Moreolo, “Performance Analysis of DHT-Based Optical OFDM Using Large-Size Constellations in AWGN,” IEEE Commun. Lett., vol.15, pp. 572–574, May 2011. C. Jao, S. Long, and M. Shiue, “DHT-Based OFDM System for Passband Transmission Over Frequency-Selective Channel,” IEEE Signal Process. Lett., vol. 17, pp. 699–702, Aug. 2010. S.Wang, S. Zhu, and G. Zhang, “Walsh-Hadamard coded spectral

[23]

[24]

efficient full frequency diversity OFDM system,” IEEE Trans. Commun., vol. 58, no. 1, pp. 28–34, Jan. 2010. S. Boussakta and A. G. J. Holt, “Fast algorithm for calculation of both Walsh-Hadamard and Fourier transforms (FWFTs),” IEE Electron. Lett., vol. 25, no.20, pp. 1352–1354, Sep. 1989. S. S. Kelkar, L. L. Grigsby, and J. Langsnery, “An extension of parseval’s theorem and its use in calculating transient energy in the frequency domain,” IEEE Trans. on Industrial Electronics,, vol. IE-30, no. 1, pp. 42–45, Feb. 1983. G. Robinson, “Logical convolution and discrete Walsh and Fourier power spectra,” IEEE Trans. Audio Electro acoustic., vol. 20, no. 4, pp. 271–280, Oct. 1972. Sivanagaraju, V., Siddaiah, P., A Robust Hybrid NDA Estimation Technique for SNR Estimation in OFDM System, (2015) International Journal on Communications Antenna and Propagation (IRECAP), 5 (2), pp. 63-69. Anoh, K., Mapoka, T., Abd-Alhameed, R., Ochonogor, O., Jones, S., On the Application of Raised-Cosine Wavelets for Multicarrier Systems Design, (2014) International Journal on Communications Antenna and Propagation (IRECAP), 4 (4), pp. 143-150. Belkadid, J., Benbassou, A., El Ghzaoui, M., PAPR Reduction in CE-OFDM System for Numerical Transmission via PLC Channel, (2013) International Journal on Communications Antenna and Propagation (IRECAP), 3 (5), pp. 267-272. Anoh, K., Ghazaany, T., Hussaini, A., Abd-Alhameed, R., Jones, S., Rodriguez, J., An Evaluation of Coded Wavelet for Multicarrier Modulation with OFDM, (2013) International Journal on Communications Antenna and Propagation (IRECAP), 3 (2), pp. 83-89. Gauni, S., Kumar, R., Sankardhayalan, G., Efficient Power Allocation in Turbo Decoded OFDM Systems Using Logarithmic Maximum A-Posteriori Algorithm and Soft Output Viterbi Algorithm Under Various Fading Environments, (2013) International Review on Modelling and Simulations (IREMOS), 6 (3), pp. 966-975. Mattera, D., Tanda, M., Data-aided synchronization for OFDM/OQAM systems, (2012) Signal Processing, 92 (9), pp. 2284-2292. Mattera, D., Tanda, M., Preamble-based synchronization for OFDM/OQAM systems, (2011) European Signal Processing Conference, pp. 1598-1602.


Research Scholar, St.Peter’s University.

2

Professor, R.M.K College of Engineering and Technology. N. Bharathi Raja received his B.E. degree in Electronics and Communicaiton Engineering (ECE) from Madras University, Chennai, India and his M.E degree from Annamalai University,India in 2000 and 2004, respectively. His research Interests are in the areas of wireless communication systems, signal and multicarrier and spread specturm Techniques.

Dr. N. Gangatharan received the B.E. Degree in Electronics and Communication Engineering in 1988 and the M.E. Degree in Microwave and Optical engineering in 1990, both from Madurai Kamaraj University. He is a Second Ranker in the M.E. Degree Examinations. He received his second M.E. Degree in Computer Science and Engineering from Manonmaniam Sundaranar University, Tirunelveli in 1997, and his M.B.A. Degree from Madurai Kamaraj University in 1999. His areas of research interest include twodimensional and multidimensional digital filters and digital signal processing.



557


Agile Development with 3D Data Modeling Kristian Sestak1, Zdenek Havlice2 Abstract – This paper deals with design of the agile method RISVICO. The method is for more effective collaboration between software engineer and database analyst by using an effective visualization tool in 3D environment, database analyst and software engineer belong to the software engineering. The method is for more effective collaboration between software and database development by using an effective visualization tool in 3D environment, database analyst and software engineer belongs to the software engineering. The visualization tool will provide important information for a database analyst and for a software engineer. Main idea is make use of nature of data which the program uses, for development and maintenance in software life cycle. Data which software will use determine procedure in development and maintenance, in this way we can provide better quality of software product. Copyright © 2015 Praise Worthy Prize S.r.l. All rights reserved.

Keywords: Software, Development, MDD, Agile, RISVICO, Software Life Cycle, Visualization Tool

We achieve got better time (shorter time for analysis and design) compared to the previous method (we have used various agile methods) which we have used), but we do not present the results yet, because we want to get relevant results.It needs to be tested for more time and we have to finish the visualization tool which is not fully completed. Our contribution should bring more effective and better defined communication between software engineer and database analyst in development and maintenance of software (software system) and also bring a mechanism of dependence between data which the software uses and development method in software life cycle. The remainder of this paper is organized as follows: section II provides an overview of software life cycle and basic principles of agile development. Section 3 is the main section in which we explain and describe main principles of our agile method for developing software. In the last section we evaluate our method and purposes for future work, which is about finishing of the visualization tool.

Nomenclature RISVICO 3D ARSA ERD MDA MDD XML

Risk and Statistics Visualization for Collaboration Three dimensional Auto-Reflexive Software Architecture Entity Relationship Diagram Model-driven architecture Model Driven Development eXtensible Markup Language

I.

Introduction

We are trying to develop software with some quality. The quality of software product should be as high as possible. There are many methods, methodologies how to achieve the quality. Methods describe several procedures, several steps. But it is still based on a general model of the software development. It means that we do all steps by the specific method without taking into account nature of the software. Because every software has smaller or bigger differences which should be adapted for the specific software product. Our solution is based the nature of data which the program uses and these data for us logically define the procedure how we should proceed in the development of the software. Our method uses visualization tool for visualizing statistics from database (for database analytics can provide useful information) which is used by the software and also the visualization tool provides data from database in specific category (for software developer can provide useful information).

II.

Software Life Cycle

There are many different models of software development life cycle. These models are referred to as process models of software development. Each process model is based on a series of steps, which are characteristic of the model and intended to ensure the success of the software development process. The software process is split arranged set of steps leading to the creation or modification of a software product [1]. All models share some basic activities, they are [2]:


558

K. Sestak, Z. Havlice

RISVICO method (see Fig. 1) consists of the following steps (a detailed description is gradually describes in the other parts of the article): 1. Evaluation of the risk of database (s) on a risk scale, the risk may be marginal, critical or very critical. There is a number of possible evaluation of a table of the database, these can take the values as marginal, critical and very critical (see Fig. 2). 2. In the source code of the program (software) is to allocate parts which use data from the database. These parts are categorized similar as in the part of the database. Then each of risk category corresponds to a certain type of procedure in the development or maintenance (see Table I). 3. Analysis of statistical data that have been created by the software product in use (helping to streamline database structure) these data will be provided by a visualization tool.

 Specification - defines the particular features and limitations of the system.  Design and implementation - the aim is the creation of a system that meets the specification.  Validation of software - the software must be tested to demonstrate that meets the requirements of the client (customer).  Evolution of software - the software must be developed so that it must be able to satisfy the customer needs in the event of changes. Furthermore, the software life cycle includes other activities such as, project management, analysis, design, implementation, quality assurance and maintenance. II.1.

Agile Methods

Agile methods are subset of iterative and evolutionary methods. They are based on iterative enhancement and opportunistic development processes [3] - [6]. Requirements for the fast introduction of software systems require changes in the methodologies. Traditional methodologiescease to functionality in such conditions and begin to explore the methodologies that allow the solution to quickly and flexibly adapt to changing requirements [7]. Every agile methodology is their specific way, but they are all built on the same principles and values [8]. That means that every agile methodology has or shares the same idea that accepts changing requirements [9]. In 2001 took place meet representatives of this approach and described manifesto agile software development, which has declared four values, and preferences [8]: 1. individuals and interactions, over processes and tools, 2. working software over comprehensive documentation, 3. customer collaboration over contract negotiation, 4. responding to change over following a plan.

Fig. 1. The RISVICO method

III. Method RISVICO The proposed method RISVICO (Risk and Statistics Visualization for Collaboration) puts the main emphasis on visualizing statistics and risk parts of the database or databases, for cooperation between database analysts and software engineers using the visualization tool. It is the agile method that allows fixed input parameters (duration of individual processes, risk levels), which need to be adjusted and adapted to the specifications of the software product and development team (e.g. number of people). The RISVICO method is based on the agile approach MDD [10] (Model Driven Development) where the MDD approach to software development method uses models that are independent of the platform to be used, and thereby increase abstraction and reduce dependence on source code dependent on a specific platform. This is a model-oriented software development, create a model (s) from which the generated source code (s) for a particular platform, then the generated code we make the specific details [11].

Fig. 2. Possible evaluation of the table of a database

III.1. Communication and Cooperation The development team have been developing software which is tested by client and after validation is created software product (software product version).



559


It can be then extended or modified during maintenance, thus forming the next version of the software product (see Fig. 3 and Fig. 4). Length of the iteration is set to 20 days then tests of software functionality and database statistics were set to 10 days (We have decided for these configurations based on our experiences and nature of the software project, for one database analyst and two software engineers) see Fig. 4.

The software product is a software system (software, program) that is fully functional and useful, has been validated by client after successful testing of functionality. Software product is identified by version and can be further adjusted based on the other client requirements, so that there is a possibility to create other versions of the software product. Each of the iteration of the method is creating a fully functional software product. Where one iteration consists of: 1. The specific requirements for application and for architecture. 2. The requirements analysis and high-level proposal. A detailed design. Code generation from UML models, integration tests. 3. The testing of the system, optionally subsystem. III.3. The Main Principles of the RISVICO Method The main idea is use nature of data which are used by the program (software) and better and more sophisticated support communication between a software engineer and a database analyst. The use of nature of the data is based on finding a good, effective and logical connection between development of software and data which the software uses (will use). More precisely, data which the software uses in the specific parts of the software should create construction based on nature of the data. First what we need to do is to divide data into some logical parts; it is not simple task, because there are can exist many divided parts which are not useful. We have decided to divide them about the risk of data. It means that the meaning of the data have different importance to the organization which uses the program. We have divided the data into these categories (see Fig. 6, Fig. 7):  Marginal - everywhere in the program source code where is this specific category of data, it will use standard procedures in the software development, it is defined in Table I.  Critical - everywhere in the program source code where is this specific category of data, it will use detailed procedures in the software development it is defined in Table I.  Very critical - everywhere in the program source code where is this specific category of data, it will use full procedures in the software development, it is defined in Table I. For example: Consider a program P, program uses two data tables T1 and T2. We will include a table T1 into category C1 and T2 into category C2. Category C1 tells us that if we use T1 in the P and the place in the code we do pre-defined steps S1. The similar situation is for category C2 tells us that if we use T2 in the P and the place in the code we do pre-defined steps S2. S1 - in the code add an error logging to a file. S2 - it needs to be used for errors, exceptions and do not use the log file.

Fig. 3. Actors and the method RISVICO

Fig. 4. Development and maintenance of software of method RISVICO

III.2. Time Length of Individual Processes in the Method and the Definition of Iteration The duration of each process shall be determined on the basis of the specifications of the software in the early stages of the development or maintenance; these values also depend on the number of people in the team. It means that used values are variables (see Fig. 5).

Fig. 5. The RISVICO method



560


TABLE I PROCEDURES OF THE METHOD RISVICO IN THE DEVELOPMENT AND MAINTENANCE OF SOFTWARE Name of Specification Requirements procedures standard Development and Standard rules must be maintenance of software followed for the system or part of the programming software system that is language.They must marginal in terms of follow standard rules for risk. the programming language. detailed Development and The same as standard and maintenance of software in addition must be used system or part of the to solve error exceptions. software system that is critical in terms of risk. full Development and The same as detailed and maintenance of software in addition, it has to be system or part of the used a log file. Each software system that is method must have a very critical in terms of detailed description. risk.

III.4. Artifacts, Actors and Activity of the Method RISVICO The RISVICO method used to visualize risk of database (s) by a visualization tool. Risk parts of the database tables are assessed according to risk data, from a certain point of view, which is determined by the organization. Artifact method consists of relational database and the number of databases which may be more than one (see Table II). Software which has been developed, or is in the process of maintenance and the visualization tool that visualizes the risk of database (s) under which then each of the logical source code applies procedures to ensure the quality and effectiveness of development that are that that the important part is given adequate attention and the less important parts are spending less effort and thus this process takes less time and ensure the quality of the software product. TABLE II ARTIFACTS OF THE METHOD

Thus, we can create any categories and to apply the specific procedure (see Fig. 7), category and the specific procedure used in the development of program code which must be in a logical connection to a given procedure.

Name of artifact Database

Specification

-

Software Visualization tool

Model of relation database

Requirements Number of tables >1, the table can be classified into categories of importance for the organization. Use a three-value scale. Structured Visualization of the model and the essential properties and statistics and risk section.

Database Analyst develops a database based on the functional requirements of the client, using a visualization tool by which analyzes statistical data from the database (s) which is already running on the basis of this analysis, then it can create more effective database structure. Software engineer develops software according to customer's functional requirements and in the parts of the source code are applied defined procedures, and also procedures determined by the visualization tool. Domain specialist together with database analyst rate on risky parts of the database tables of data. Project manager is responsible for planning and budget. Client specifies requirements for a software product which is tested and validated, and if the application meets all the functional requirements the software product is designed with a defined version (see Table III). After one iteration of the software, it becomes the software product, provided that the software has been validated by client (see Fig. 8 and Table IV). The communication between actors is in the oral, written and electronic form. In the third case is the use of different electronic communications services and tools such as email, forums and other communications tools. Choice of tools and services for communication is not particularly limited, and it is up to the actors to what was agreed before the communication takes places as the most effective.

Fig. 6. Application procedures for individual parts of the source code

Fig. 7. An Evaluated database system and its application to a software system



561


Name of actor Database analyst

Software engineer

Domain specialist

Project manager Client

TABLE III ACTORS OF THE METHOD Specification Requirements Database Number of people > 0, development knowledge of the belongs to the relational database. development team. Development and Number of person > 0, maintenance of knowledge of structured software, belongs to or object oriented the development software. team. Expert in the field in which the software product is used. Planning and Number of people 1. budget. Number of people > 0, testing the functionality of the software product.

Title of activity Iteration

Communication

Testing

TABLE IV ACTIVITIES OF THE METHOD Specification Requirements The each iteration, Functional software (part one cycle is set to a of) the software that can length of 20 days. be deployed to the real operation. There is electronic, Oral communication or written or oral. through electronic services and tools. It tests software or a When testing was software product. successful, as expected, and there are no additional requirements.

Functionality testing is conducted by the customer, once all the functional requirements and validation testing has been completed software becomes the software product. The software product may or may not continue in testing, it depends on specifications and other factors. III.5. The Proposed Tool of the Visualization One of the important things of RISVICO is the visualization tool for communication between database analyst and software engineer (see Fig. 9 and Fig. 10). Main purpose of visualization is to make an interactive visual representation of information, which influences human perception and cognitive abilities in solving the problems. Possibilities of interaction are very important, but how we will find out later, they also bring many problems. They bring into the visualization process some part of subjectivity that means comparing and measuring of efficiency and others things which are much more complicated [12]. The visualization tool will base on the ARSA [13]. In Fig. 11 is indicated method of using selfreflexive knowledge at the source level for process of compilation of the visualization tool. Our visualization tool is implemented with layer of knowledge. The layer contains: 1) Complete Entity Relationship Diagram (ERD). 2) Defined rules for ensure the referential and domain integrity. 3) Current range of tables (number of rows and columns, range of used space). 4) Number of queries. 5) Frequency of changes in parts of database. 6) Other properties. 1 and 2 - Gained from the design process and maintained consistently by automatically implementation via Model-driven architecture (MDA). They are indispensable for correct modification and expansion of the system 3 - Gained by XML file.

Fig. 8. One iteration of the software product set to the length of 20 days

The input XML file will contain entities and their relations, risk attributes and statistics, where by syntax analysis (parsing) is the process of analyzing the sequence of formal elements in order to determine their grammatical structure to the already ahead of the formal grammar (see Fig. 12). The data will be divided into pre-defined structures, structures for the programs and structures of entities and the number of entities can be in many relationships. Subsequently, by using the transformation rules, and different algorithms transformed data tables to visual structures we will adapt a specific context, in this case a three-dimensional environment (see Fig. 12). Visual structures with the help of user interaction reflect the 3D environment; user then has the possibility to influence the final visualization by creating interaction with visualization tools and so can more effectively show for him important information (see Fig. 12).



562


Fig. 11. Process of compiling of the visualization tool based on ARSA

Fig. 9. Database analyst

Fig. 12. Architecture of the visualization tool

Description of main architectonic block of the visualization tool (see Fig. 12):  Data representation: the data items which to be visualized, usually stored in a database.  Data abstraction: there are elements will be visualized (tables and relations).  Graphical objects mapping: creating visual tables and visual association between the tables (visual relations).  Visualization: graphical data presentation and the interaction with the user. Interaction: contains all related thing to the user interaction, scaling, item selections, colors associations, set up colors, critical values. Rotations around axis x, y, z. The visualization tool consists of the following layers (see Fig. 13):  Knowledge Layer - includes knowledge domain (area where the tool works) required to implement.  Implementation layer - implements knowledge domain in particular environment.  Visualization model layer - a visual 3D interactive model.  Environment knowledge layer - specific skills for specific environments such as operating system, 3D interface and others.

Fig. 10. Software engineer



563


[2]

[3] [4]

[5]

[6]

[7]

[8] Fig. 13. Layers of the visualization tool

IV.

[9]

Conclusion

[10]

The RISVICO method is particularly suitable for software systems, in which is not known in advance how the software system will look like; there are mere partial or a few specified requirements. Another condition is the use of database system (database), and also that the database system can be divided into different levels of importance depending on the potential negative impacts (in case of failure, malfunction, etc.) for the organization that uses the software. Or another system can be divided in a manner, such that the more important parts will be given more attention and will be addressed in more detail then the less important parts using standard procedures, and in this way we can achieve a better quality and save time. Division of the database system may not be about risk, however, it is sufficient if we divide the data into categories and specific category corresponds to a specific procedure in the software (program, software system). In other words, each of the plurality of possible types (not here classics data such as integer, string, etc.) Given in the database must have its defined procedures in a software system. In this way, we can create then a more efficient structure (in terms of the use of the right structures in defined parts of) the software system (program, software), if we find the right tips that will distribute data in the database (database system). We presented only one of many possible options. Our work will be further about finishing the visualization tool and the RISVICO method should be further analyzed and tested for confirmation the present results.

[11]

[12]

[13]

[14]

[15]

[16]


Steelcon Slovakia s.r.o., Južnátrieda 1598/82, 040 17 Košice, Slovakia. E-mail: [email protected] 2

Technical University of Košice, Faculty of Electrical Engineering and Informatics, Department of Computers and Informatics.Letná 9, 04200 Košice, Slovakia. E-mail: [email protected] Kristián Šesták was born on 10.12.1978. In 2006 he graduated (MSc.) at the Department of Computers and Informatics of the Faculty of Electrical Engineering and Informatics at Technical University in Košice. Hi is currently studying his PhD his scientific research is focused on an effective modeling software in 3D.

References [1]

nyrstvi.pdf>, Ostrava 2002, 4. Softwarovémetriky vykazovani-vyuky-cvut, , 5. Larman, C., Agile and Iterative Development: A Manager's Guide. Boston: Addison Wesley, 2004., 10. Larman, C., and Basili, V., "A History of Iterative and Incremental Development," IEEE Computer, vol. 36, no. 6, pp. 47-56, June 2003., 11. Basili, V., R., and Turner, A., J., "Iterative Enhancement: A Practical Technique for Software Development," IEEE Transactions on Software Engineering, vol. 1, no. 4, pp. 266 270, 1975., 12. Curtis, B., "Three Problems Overcome with Behavioral Models of the Software Development Process (Panel)," International Conference on Software Engineering, Pittsburgh, PA, 1989, pp. 398-399., 13. Buchalcevová, A., Metodikybudováníinformačníchsystémů, kategorizace, agilnímetodiky, vzory pro návrhmetodiky, 1. vyd. Praha: Grada, 2005. 163 s. Management v informačníspolečnosti. ISBN 80247-1075-7.,2. Fowler, M– Highsmith, J.: The Agile Manifesto, Software Development, August 2001.,1. The agile development, , 3. Selic, B., The Pragmatics of Model-Driven Development, Published by the IEEE Computer Society, IEEE SOFTWARE, 2003,15. Pastor, O., España, S., Panach, J., I., Aquino, N., Model-Driven Development, Informatik-Spektrum , Volume 31, Issue 5 , pp 394407, ISSN 0170-6012, 2008,14. Khan, M., Khan, S., S., Data and Information Visualization Methods, and Interactive Mechanisms: A Survey, International Journal of Computer Applications (0975 – 8887), Volume 34– No.1, November 2011,17. Havlice, Z., Auto-reflexive software architecture with layer of knowledge based on UML models, (2013) International Review on Computers and Software (IRECOS), 8 (8), pp. 1814-1821. Šesták, K., - Havlice, Z., Visualization of Critical Properties of Databases of Information Systems, SAMI 2015: IEEE 13th International Symposium on Applied Machine Intelligence and Informatics: proceedings: January 22-24, 2015, Herľany, Slovakia. - Košice, 2015. Šesták, K., - Havlice, Z., Úvod do vizualizáciesoftvéru, In: Datakon 2014: Proceedings of the 34th Annual Database Conference: Demänovskádolina, Jasná, Slovak Republic, September 25-27, 2014. Šesták, K., VYUŽITIE VIZUÁLNEJ REPREZENTÁCIE MODELOV V POSTPROJEKTOVÝCH FÁZACH ŽIVOTNÉHO CYKLU SOFTVÉROVÝCH SYSTÉMOV, Dizertačnápráca, 2015.

Vondrák, I., Úvod do softwarovéhoinženýrství, verze 1.1, VŠB – Technickáuniverzita Ostrava, Fakultaelektrotechniky a informatiky, katedrainformatiky, = L. We can use Eq. (6) to decide the best Sb:

2

Proposed Watermarking Approach

We have implemented DCT watermarking technique for embedding the region of interest for an image into the carrier image; ROI is the fundamental portion of an image that can completely identify an image. The algorithm has two phases, watermark embedding and watermark extraction.

2

(1)

Step 5: Get the next block B(b,d), where b, d are the indices of B’s first pixel located at the upper left corner of B. b, d is incremented by 8 in each iteration. Step 6: Apply DCT transform on B to obtain Bt which has the same size as B.



644

Nadhir Ben Halima, Osama Hosam

Step 7: Convert Btto vector V. Step 8: Get the median value of V. When V length is odd the formula is:

II.2.

DCT Watermark Extraction

The extraction has been done with the same procedure of embedding except for the embedding steps; the original image is divided into 8 × 8 non-overlapping blocks.

(2) when V length is even the formula is:

2

(3)

Step 9: Repeat Step 5 to Step 8 for each block, when finished put all the median values into single vector M. Note that M length is Nb Step 10: Go to step 3 to get the next Wi select the nearest value to Wi in the vector M. This will be decided according to the following formula: (d, p)=min(|M-Wi|)

(4)

where d is the resulting minimum difference and p is the position of the minimum difference. Step 11: Change M value at position p with the watermark value Wi. Put p and i index pair into K vector each pair is an element in the vector. K is the Key used in the extraction procedure. K length is L. Step 12: Repeat Step 10 and Step 11 until i= L. Get the inverse of all blocks after changing their median values.

Fig. 2. DCT Watermark Extraction

DCT transform will be applied on each block; the median value of the transformed blocks will be obtained and put into a vector M. The extraction key K will be used to restore back the watermark values and sort them back to obtain the transformed watermark W. Finally the inverse DCT is applied on W to obtain the watermark or the region of interest ROI. The following procedure, shown in Fig. 2 is used in the extraction process. Step 1: Divide the image Is (the watermarked image) of size n × m into 8 × 8 non-overlapping blocks. The block size Sb can be decided according to the watermark length L. So the number of blocks Nb >= L. use equation (2) if a block size different than 8 × 8 is selected in the embedding procedure. Step 2: Get the next block B(b,d), where b, d are the indices of B’s first pixel located at the upper left corner of B. b, d is incremented by 8 in each iteration. Step 3: Apply DCT transform on B to obtain Bt which has the same size as B. Step 4: Convert Bt to vector V.

Fig. 1. DCT watermarking



645


Step 5: Get the median value of V. Use Eqs. (2), (3). Step 6: Go to Step 2, iterate number of steps = Nb, when finished put all the median values into single vector M. M length is Nb. Step 7: Use K’s p to extract the vector W’ which is the watermark values not arranged in the true order, the length of the vector K is L. Use K’s i to sort the extracted values and obtain W. W vector must be reshaped to square matrix of size r × c by using the following formula: (5) √

10

where MAXI is the maximum intensity value of the image pixels. MSE and PSNR have an inverse relationship, i.e large values of PSNR mean high quality of embedding and high similarity between the compared images. While large values of MSE mean lower similarity and high degradation of the watermarked image compared to the original image. This comes from the fact that PSNR measures a ratio between the signal and noise or S/N where S is the signal and N is the noise. So large values of S mean large value of the ratio and small values of Noise lead also to large values of the ratio. PSNR is measured in units of decibels (dB). PSNR with a value of 40 dB means that the watermarked image and original image are indistinguishable by Human Visual System. Typically most watermark approaches get embedding quality ranges from 30 to 40 dB’s. To evaluate the watermarking technique the following equation is used:

Step 8: Apply inverse DCT on W to obtain R.

III. Results and Discussion The evaluation of the watermarking process is done according to the quality of the extracted watermark and the watermarked image. Using the quality measures, robustness of the watermarking procedure can be evaluated especially when the watermarked image is attacked. The attack can differ from adding noise to trying to remove the watermark from the watermarked image. The main objective is to make sure that the quality of the image after watermarking is still acceptable. To evaluate the quality of the watermarked image we introduced MSE, PSNR, SF and SSIM [7] We are introducing here the most common measures used to compare the quality of images in watermarking approaches; namely the Mean Square Error (MSE) and the Peak Signal to Noise Ratio (PSNR). MSE defines the image as collection of numbers contained in a matrix and calculates the accumulated average of squared error measured between the original image and the watermarked image. PSNR is a measure of the peak error. MSE gives estimation of the difference between both images. MSE of 0 means no difference between the original image and the distorted image. Given an image I (i, j) and an image K (i, j) of equal dimensions, the MSE is defined to be: 1

,

,

(7)

∑ ∑

∑

∑

, ,

(8)

,

or: ∑

10

∑

∑

∑

, ,

,

(9)

are original image and watermarked image where , respectively. The Structural SIMilarity (SSIM) [9] index is a method for measuring the similarity between two images. For calculating SSIM, the quality measure is considered between two images assuming one of them in perfect condition. SSIM proposed extracting the luminance from the scene since it doesn’t affect the object structure in the image. The difference between SSIM and other approaches such as MSE and PSNR is that these approaches concentrate on calculating the perceived error; but SSIM considers the error as the perceived degradation in the structural information of the objects in the scene. Structural information is considering the interdependence of pixels inside the image which carry very important information about the structure of the object in the visual scene. SSIM is measured using the following formula:

(6)

where M and N are the dimensions of the image. MSE is often used as a measure between two images because it measures the energy of the difference between both images. This will simplify calculations of typical MSE since image processing transforms are typically energy preserving [8]. Another common performance measure used frequently in image processing is PSNR. It measures the ratio between the maximum possible power of the signal and the possible percentage perturbation of the signal or the added noise. The quality of the original image I(i,j) with N × M pixels and the watermarked images K(i,j) with N × M pixels are calculated using the following formula:

,

2

1 2 1

2 2

(10)

where µx the average of x, µy the average of y, σ2x the variance of x, σ2y the variance of y, σxy the covariance of x and y, c1=(k1L)2, c2=(k2L)2 are two variables to stabilize the division with weak denominator, L the dynamic range of the pixel-values (typically this is 2#bits per pixel−1) and k1=0.01 and k2=0.03 by default.



646


An example of block DCT sorted coefficients

1.2835 0.0056 0.0048 0.0046 0.0032 0.0006 0.0002 0.0002 The resulting images after embedding in highest, second, sixth and lowest DCT values of the selected blocks in the image

(a)

(b)

(c)

(d)

Figs. 3.The effect of embedding into different locations of DCT coefficients

affect the block’s image. But to make minimum perturbation or distortion of the entire image, each watermark Wi value is compared with a vector M which contains all the median DCT values of the blocks. Wi will be embedded in the coefficient with minimum distance as: [D, P] = min(|M-Wi|) (11)

III.1. The Location of the Carrier Blocks The DCT coefficients are perturbed for embedding the watermark. First the watermark is embedded into the lowest frequency coefficient values of each selected block of the image. For example, suppose the following DCT coefficients are given [1.2835 0.0048 0.0032 0.0002 0.0002 -0.0006 -0.0046 0.0056]. The absolute values of the above DCT coefficients are sorted in descending order, we obtain 1.2835 0.0056 0.0048 0.0046 0.0032 0.0006 0.0002 0.0002. The first value is the DC value or 0 frequency value. It is the highest DCT value. Embedding the watermark into the DC value of all blocks will distort the image as shown in Fig. 3(a). This occurs because we changed the base frequency which will affect the entire image block. This appears when, first; the block is extracted from the original image, then it is transformed by using DCT, then it is perturbed by changing the DC value, and finally the inverse transform is obtained for the block. The resulting image block values are completely different from the original image block values. When the second DCT value is used for embedding, the resulting image has blocky effect as shown in Fig. 3(b). It should be noted that the first value of the watermark is the DC value so it is usually high value and will affect the entire image after embedding. Its effect will completely change the block shape as shown in Fig. 1 at the upper left corner of each image. When the minimum DCT values are used in embedding, the resulting image has a blocky effect. So embedding must be done around or in the middle of the blocks’ DCT values. Now, if all the middle coefficients of all blocks are collected, it will be better not to embed the watermark serially starting from the first block. Instead we must search for the most suitable block for embedding as shown in the following procedure. The image is divided into 8 by 8 blocks; each block is transformed by using DCT transform. The resulting DCT coefficients for each block are sorted in descending order. Then embedding is done only in the median coefficient of each block, as stated before, it will not

where D is the distance value of the median DCT coefficient with minimum distance between itself and the watermark value Wi, P contains the position of that coefficient. It should be noted that, not all the blocks will be used for embedding unless there is a watermark vector longer in length than the number of blocks in an image. Figs. 4 show the blocks used for embedding a portion of the image (ROI) with size 50 × 50. The watermark contains 2500 DCT coefficients. The embedded watermark coefficients are manually multiplied by 1000 to distort the embedding blocks to be visible for analysis purposes. From the figure we notice that the algorithm selected the blocks with high frequency components. This appears in the image as a collection of blocks around vegetable’s borders. The watermark extraction is executed by searching for the watermarked blocks. The first challenge was to find the order of the watermark values w1, w2, Etc. Two loops are used for navigating through the image to extract the watermark. The problem is that the two loops navigate the image serially. So the search starts by the first upper left block, then the second block to the right and so on. For example Fig. 4(a) extracts the watermark in the following order (w2, wn, w3, w4, w1) which is not the actual watermark values order. So, the watermark block index must be added to the key in the extraction procedure. After extracting the watermark the values are rearranged according to the watermark index. III.2. Embedding Using Scaling Factor α A collection of 200 images 512 × 512 pixels are collected as a dataset for the experiments.



647


The resulting images with the extracted watermarks are shown in Figs. 5. The PSNRs are 34.24, 39.85, 37.15, and 36.86 dBs for Pepper, Lake, Walkbridge and Pirate watermarked images respectively. PSNR values are higher in images with high variance in intensities such as Lake and Walkbridge. The variance in intensities occurs because of the existence of trees and versatile background. Visible distortions in the watermarked image are encountered due to embedding high coefficient values such as DC value. Such problem can be avoided if the coefficients are controlled by scaling factors α.

For embedding with the above procedure we selected ROI as 50 × 50 pixels. ROI size will be fixed through all experiments for better comparison. DCT transform for ROI is obtained; the resulting coefficients are 50×50=2500 coefficients. The coefficients are embedded into the blocks with best median values. The dataset images are 512 × 512 grayscale with 8 bits color depth. The image is divided into 8 × 8 blocks to obtain 64 × 64 blocks or 4096 blocks, i.e. we can embed DCT coefficients up to 4096 coefficients in each image. ROI size can be selected maximally to be 64 × 64, so each image block can hold single DCT coefficient.

(a)

(b)

Figs. 4. Embedding by using sorted DCT coefficients and selecting the most suitable block for each watermark value. (a) The selection criteria is done according to the relative difference between the watermark value and the minimum value of DCT coefficients of all blocks, arrows indicates the locations of each watermark value(b)An example shows the embedding of ROI in the suitable blocks, it is manually distorted for analysis purposes

(a) Pepper

(b) Lake

(c) Walkbridge

(d) Pirate

Figs. 5. The results of median based DCT watermarking. Rows represent from top to bottom, Original images, Watermarked images and the extracted and embedded watermarks. Arrows point to visible distortion in the image.(a) Pepper image (b) Lake Image (c) Walkbridge image (c) Pirate image



648


(a) Original image

(b) Watermarked image, PSNR = 34.25

(c) Watermarked Image c=0.2, PSNR = 44.42

Figs. 6. Using scaling factor in embedding (a) the Original Image (b) The watermarked image without using the scaling factor, arrows point to visual distortion (c) The watermarked image with using α=0.2

For Lake, Walkbridge and pirate shown in Figs. 7(b), (c) and (d) respectively, similar curves are obtained. Scaling factor α is different for each image (not too much different). For example, the best α value is 0.1 for Lake Image, 0.3 for Walkbridge image and 0.15 for Pirate image.

In embedding procedure, the watermark DCT coefficients are multiplied by the scaling factor. In extraction procedure, the extracted watermark DCT coefficients are divided by the scaling factor so the original watermark is obtained. Figs. 6 show the pepper image with embedding 50 × 50 ROI image into it. When the parameter α is not used, i.e. α=1, the watermarked image is distorted with noticeable distortion. When using the scaling factor, α = 0.2, the watermarked image is clear from distortion. HVS can’t discriminate between the original image and the watermarked image. In addition PSNR is raised from 34.25 to 44.42 dB. The watermarking is done into the median DCT coefficients. This can be translated into the medium frequency bands in the image. Middle frequency values between the highest frequency in the image and the lowest frequency in the image. The highest frequency areas in the image are the areas with high intensity variations. While the low frequency areas in the image are the areas with backgrounds or no intensity variations. What is the best value of α? Raising α affects negatively on the image quality, since greater values will affect the carrier image and make visual distortions. Reducing α will decrease visual perturbation and increase the watermarked image quality appearing in the higher values of PSNR. On the other hand increasing α strengthen the existence of the watermark. The watermark is visually intact after extraction when embedding is done with higher α value. For lower α values, the watermark is distorted. As a result we have a trade-off of using α parameter. α value can’t be increased too much or decreased too much. Increasing it too much affects the image quality and decreasing it too much affects the watermark robustness. Fig. 7(a) shows the plot of PSNR of the Pepper watermarked image and SSIM of the extracted watermark for different values of α. PSNR ranges from 55.34 for α=0.01 to 39.47 for α=1. SSIM starting value is 0.02 for α=0.01. SSIM gradually increases and give the highest value which is 0.5 with α=0.3 afterwards SSIM gradually decreases until reaching 0.2 with α= 1.

III.3. Robustness to Attacks When SSIM = 1 then 100 % of the watermark is extracted, also when SSIM =0.37 then 37% of the watermark can be extracted. Table I shows the robustness of the proposed approach with different attacks. For Gaussian noise, the watermarking procedure is robust; approximately 40 % of the watermark can be reconstructed and recognized by HVS. For Salt & Pepper noise, when percentage of noise is added to the carrier image using the same percentage such that used in Gaussian noise, the watermarking procedure is more robust compared to Gaussian noise. So if 40 % of the watermark can be extracted with Gaussian noise, we can extract 68 % of the watermark with the same amount of Salt & Pepper noise. In JPEG compression we used different percentage, 30% refers to the percentage of compression, in this case 30% of compression refers to 70 % image quality. It appears from Table I that with JPEG quality higher than 90 % the embedded watermark can be extracted completely. With Multiplicative noise, if the same amount of noise is added as for Gaussian Noise and Salt & Pepper noise, then the proposed procedure is more robust to Multiplicative noise than Gaussian Noise and less robust to Multiplicative noise compared with Salt & Pepper noise. If 40 % of the watermark can be extracted with Gaussian noise, then we can extract 68 % of the watermark with the same amount of Salt & Pepper noise and also we can extract 52 % of the watermark with the same amount of Multiplicative noise. For Histogram equalization and image un-sharp attacks, 58% and 63% of the watermark can be extracted respectively. The watermark can still be recognized with HVS. The proposed watermarking procedure is robust to geometrical attacks such as cropping and rotation.



649


IV.

(a) Pepper

Novel digital image watermarking approach has been proposed. The watermarking procedure is mainly proposed to embed ROI part of the image into the image itself. The watermark is embedded into the median DCT coefficients of 8 × 8 blocks of the image. The median coefficients for all blocks are collected and a comparison is done to find the best median coefficient for watermark embedding. The results showed that embedding into the median coefficient is not affecting the whole quality of the watermarked image. In addition embedding in median DCT coefficients is more robust to attacks compared to lower and higher DCT coefficients.

(b) Lake

(c) Walkbridge

References

(d) Pirate [1]

Figs. 7. The effect of using the scaling factor α on the watermarked image quality (measured by PSNR) and the extracted watermark quality (measured by SSIM) (a) Pepper image (b) Lake image (c) Walkbridge (d) Pirate TABLE I EMBEDDING INTO INDEX 8, TDOFF=10, OF F16 IMAGE, WITH DIFFERENT ATTACKS

[2]

Attack Name

Percentage

PSNR

MSE

SSIM

Gaussian Noise

3% 1% 0.1 % 0.01 % 3% 1% 0.1 % 0.01 % 30% 10% 1% 0.1 % 3% 1% 0.1 % 0.01 % -

15.1407 19.3 28.7767 36.3 19.1321 23.8 32.4 38.1 35.2511 37.4 39.5 39.5391 17.4870 21.8416 31.2581 37.6418 10.8692

1633.3351 626.7681 70.7215 12.2480 651.6418 221.1338 30.0702 8.0899 15.9261 9.7012 5.9336 5.9336 951.7416 349.1886 39.9397 9.1841 4368.6646

0.0052 0.0949 0.3778 0.9853 0.1706 0.3419 0.6840 0.9672 0.5404 0.9992 1 1 0.0995 0.2096 0.5244 1 0.5831

-

21.8266 31.3852

350.3991 43.6328

0.6307 0.3877

-

34.5494

21.0569

0.4377

-

38.4778

8.5222

0.8554

10% 30°

13.3602 10.2632

2768.1245 5649.3502

0.7603 0.0084

Salt & Pepper

JPEG Compression

Multiplicative noise

Histogram Equalization Unsharp Filter Average Filter [3×3] Median Filter [3×3] Gaussian Low Pass Filter Cropping Rotate

Conclusion

[3]

[4]

[5]

[6]

[7]

[8]

[9]

[10]

[11]

The watermarked image is extremely distorted, but can be recognized by human visual system compared to other attacks such as Gaussian noise addition. The extraction of the whole watermark can’t be expected, however if the parameters of cropping and rotation attacks are known, the watermark or part of the watermark can be restored successfully. The cropping parameters such as location and size of cropping are obtained, then the rotation parameter such as the angle of rotation is obtained then part or the entire watermark can be clearly extracted.

[12]

[13]

Jason Dowling, Birgit M. Planitz, Anthony J. Maeder, Jiang Du, Binh Pham, Colin Boyd, Shaokang Chen, Andrew P. Bradley, and Stuart Crozier “A comparison of DCT and DWT block based watermarking on medical image quality” Proceeding IWDW '07 Proceedings of the 6th International Workshop on Digital Watermarking Pages 454 - 466 Springer-Verlag Berlin, Heidelberg ,2008 Pithiya, Pravin M., and H. L. Desai. "DCT Based Digital Image Watermarking, De-watermarking & Authentication." image 1.2: 7. Osama Hosam,“Side-Informed Image Watermarking Scheme Based on Dither Modulation in the Frequency Domain”, The Open Signal Processing Journal, 2013, Issue 5, Pages1-6. Wen Yuan Chen and Shih Yuan Huang "Digital Watermarking Using DCT Transformation. “Department of Electronic Engineering National Chin-Yi Institute of Technology (2000). Osama Hosam, ZohairMalki. Steganography Technique for Embedding Secure Data into the Image Regions with Abrupt Changes.Life Sci J 2014;11(9):126-130. Fung, Charles, AntônioGortan, and Walter Godoy Junior."A review study on image digital watermarking." ICN 2011, The Tenth International Conference on Networks. 2011. Cox, Ingemar J., et al. "Secure spread spectrum watermarking for multimedia."Image Processing, IEEE Transactions on 6.12 (1997): 1673-1687. RaziehKeshavarzian, “A New ROI and Block Based Watermarking Scheme Using DWT” 20th Iranian Conference on Electrical Engineering, (ICEE2012), May 2012. Rajkumar.V1 and Arunkumar.J.R2. "A ROI and Block Based digital image watermarking using Discrete Wavelet Transform."International Journal of Emerging Trends & Technology in Computer Science (IJETTCS) Volume 2, Issue 3, May – June 2013. DashunQue, Li Zhang, Ling Lu and Liucheng Shi, “A ROI Image Watermarking Algorithm Based on Lifting Wavelet Transform”, Proceedings of International Conference on Signal Processing, Vol. 1, No. 1, 2006. Johnson, Neil F., ZoranDuric, and SushilJajodia. Information hiding: steganography and watermarking: attacks and countermeasures. Vol. 1.Springer, 2001. Havlicek, J. P., P. C. Tay, and A. C. Bovik. "AM-FM image models: Fundamental techniques and emerging trends." Handbook of Image and Video Processing 2 (2005). Z. Wang, A. C. Bovik, H. R. Sheikh and E. P. Simoncelli, "Image quality assessment: From error visibility to structural similarity," IEEE Transactions on Image Processing, vol. 13, no. 4, pp. 600-612, Apr. 2004.

Authors’ information 1,2

The collage of Computer Science and Engineering in Yanbu, Taibah University, Saudi Arabia.



650


2 The Research City for Scientific Research and Technology Applications, IRI institute, Alexandria, Egypt.

Nadhir Ben Halima received his BSc in Computer Engineering from the National School of Computer Sciences (ENSI), Mannouba, Tunisia, the M.S degree in Communication Networks Engineering from SantAnna School of Advanced Studies, Pisa, Italy in 2006 and the PhD in Information and Communication Technology from the University of Trento, Italy, in 2009. In 2009 he was a visiting researcher at the Department of Electrical and Computer Engineering at North Carolina State University, USA. Since September 2011, he is an assistant professor at the College of Computer Science and Engineering at Taibah University , Yanbu Branch , Saoudi Arabia. His research interest include wireless and sensor networks, cognitive networks, image watermaking and multi-robot systems. Osama Hosam is Assistant Professor in Research City for Scientific Research and Technology Applications, Alexandria, Egypt. In 2007 he received his MSc. In computer systems and engineering, He pursued his study in Hunan University, China and worked in parallel in Nanjing University of Technology; in 2011 he received his PhD in Computer Engineering. He moved then to Saudi Arabia and worked and Assistant Professor and then promoted to be the Head of Computer Science department, the Collage of Computer Science and Engineering in Yanbu. His research interests include, Computer Graphics, 3D Watermarking, Stereo Vision, and Pattern Recognition.



651

1828-6011(201506)10:6;1-6 Copyright © 2015 Praise Worthy Prize S.r.l. - All rights reserved