International Journal of Computer Science and Information Security (IJCSIS), Vol. 15, No. 5, May 2017
Fish Disease Detection Using HOG and FAST Feature Descriptor 1
Shaveta Malik
Research Scholar, School of Computer Science, Lingaya’s University, Faridabad, Haryana
2
[email protected]
Tapas Kumar
Associate Dean, School of Computer Science, Lingaya’s University, Faridabad, Haryana
3
A.K Sahoo
Scientist, Biodiversity Lab, Reverie Ecology and Fisheries Division, Barrackpore, Kolkata
Abstract In the paper detect the disease of fish which is EUS (Epizootic ulcerative syndrome) through the two feature descriptor .EUS(Epizootic ulcerative syndrome) disease is still misidentified by the people and it is caused due to fungus which is Aphanomyces invadans.Correct diagnose and prévention of disease is very necessary for the growth of the nation as fish is a food of millions of people .To extract the feature and to achieve the fast and accurate region of interest (ROI) of EUS infected fish through FAST (Features from Accelerated Segment Test) and HOG (Histograms of Oriented Gradients) Feature Descriptor. The Experimentation done on real images of EUS infected fish in MATLAB environment.
Keywords:-FAST (Features from Accelerated Segment Test), EUS (Epizootic ulcerative syndrome), HOG (Histogram of Oriented Gradient), Neural Network (NN)
1. INTRODUCTION Epizootic ulcerative syndrome (EUS) is a common fish disease found in fresh water and brackish water fishes. The disease is clinically diagnosed as deep ulcers and heavy mortality caused by Aphanomyces invadans, a fungal pathogen. It is still misidentified by the people. To identify the features of EUS (Epizootic ulcerative syndrome) infected fish and finds the region of interest for the accurate and fast diagnose. The concept of feature detection refers to methods that are aim at computing abstractions of image information and making local decisions at every image point whether there is an image feature of a given type at that point or not. The resulting features will be subsets of the image domain, often in the form of isolated points, continuous curves or connected regions. Extract the features from two different descriptor then reduce the dimensionality through PCA (Principle component analysis) after that classify the EUS and Non-EUS infected fish through classifier and then compare the classification accuracy through same classifier .After the experimentation ,it has been observed that the combination FAST-PCA through classifier gives better accuracy as compare to the HOGPCA. Many researchers has done lot of work on different feature descriptor,Dalal et al.[1] extracted the HOG features from all locations of a dense grid on a image region and the combined features are classified by using linear SVM. They showed that the grids of HOG descriptors significantly outer-performed existing feature sets for human detection. Jeyanthi Suresh et al.[2] In the paper, proposed a method which automatically recognizes the human activity from input video stream with Histogram of Oriented Gradient features (HOG) and Probabilistic Neural network (PNN) classifier and In which features are extracted from input video frames by HOG features detector and are concatenated to form an action pattern. The Probabilistic Neural network (PNN) is used as a classifier for classifying the actions of supplied video. Experiments are conducted on KTH database and gives better performance in terms of 100% recognition rate for training set and 89.8% accuracy for test set. SVM classifier is used for comparison and measuring performance of each feature set with respect to two classifier.Valentin Lyubchenko et al. [3] in the paper choose or select the color markers to detect the infected area or to distinguish the normal and infected area. The main disadvantage of this methodology is the appearing of false point marked as infected areas due to the automatic allocation of the selected color which may affect the calculation. On the contrary, this methodology has many advantages such as: the ability to change the size of the
216
https://sites.google.com/site/ijcsis/ ISSN 1947-5500
International Journal of Computer Science and Information Security (IJCSIS), Vol. 15, No. 5, May 2017
marker when selecting the colors in the image segmentation to avoid any appearance of false points, the ability to analyze fish skin areas that are difficult to identify visually and the ability to isolate areas of the skin with varying levels of destruction. Edward Rosten et al.[4] In this paper, FAST family of detectors have been presented. In which turned the simple and very repeatable segment test heuristic into the FAST-9 detector using machine learning, which has unmatched processing speed. The resulting detector has excellent repeatability despite the design for speed. By removing preconceived ideas and generalizing the detector about how a corner should appear, we were able to optimize a detector directly to improve its repeatability, creating the FAST-ER detector. While still being very efficient, FAST-ER has dramatic improvements in repeatability over FAST-9 (especially in noisy images). The result is a detector which is not only computationally efficient, but also has better repeatability results and it is more consistent with variation in corner density than any other tested detector.
2. Proposed Methodology Acquired
Apply HOG
Image
Extract Features of Each Block
Apply FAST
Extract Feature through all corners
Database
Train NN(Neural Network)
Test NN(Neural Network)
Accuracy
Figure 1:-Block Diagram of the System
Step 1:Pre-Processing: Acquire the image (the experimentation has been done on real images) after that apply the operations,i.e noise removal, convert into gray scale, crop the region of intrest,applied filter(Gaussian Filter). Step 2:-Feature Extraction:-Extract the feature through HOG(Histogram of Oriented Gradient) and FAST (Features from Accelerated Segment Test),HOG extract the feature from every block through gradient calculation and then normalize it after cell Histogram and FAST extract from the corners or the interest point which has local information content. Step 3: Classification:-After extracting the features through HOG (Histogram of Oriented Gradient) and FAST (Features from Accelerated Segment Test) apply the same classification algorithm to find the classification accuracy and performance. Neural Network machine learning algorithm applied to train the dataset and to test the dataset.
217
https://sites.google.com/site/ijcsis/ ISSN 1947-5500
International Journal of Computer Science and Information Security (IJCSIS), Vol. 15, No. 5, May 2017
3. PCA(Principle component analysis) with Feature Descriptor . Pre‐ Processing
Apply HOG
Extract Features of Each Block
Apply PCA
Apply FAST
Extract Feature through all corners
Database
Train NN(Neural Network)
Test NN(Neural Network)
Accuracy
Figure 2: Block Diagram of Proposed System
If PCA (Principle component analysis) apply after extracting the features from feature descriptor then it gives better accuracy as compared to without applying PCA through classifier, Extract the feature through HOG and FAST apply PCA for dimension reduction. PCA is mostly used as a tool in exploratory data analysis and for making predictive models. PCA is be done by eigenvalue decomposition of a data covariance (or correlation) matrix or singular value decomposition of a data matrix, usually after mean centring (and normalizing or using Z-scores) the data matrix for each attribute.[5] The results of a PCA are usually discussed in terms of component scores, sometimes called factor scores (the transformed variable values corresponding to a particular data point), and loadings (the weight by which each standardized original variable should be multiplied to get the component score)[6].
3.1 HOG (Histogram of Oriented Gradient):-It is a feature extraction algorithm, in which generate features form the Histograms of the gradient orientations of the localized portion of the input image. The method divides the input image into small connected regions known as localized cells. Histograms of the edge orientations or gradient directions are compiled for each pixel within the localized cells and then this histograms are compared to represent the descriptor. The HOG Descriptors is consists of five steps:1) Image Scanning 2) Gradient Computation 3) Cell Histogram Generation 4) Block Normalization First scan the image and then apply the gradient computation and then calculate the magnitude of gradient after that apply the histogram on each cell and then finally normalize the overall block.
218
https://sites.google.com/site/ijcsis/ ISSN 1947-5500
International Journal of Computer Science and Information Security (IJCSIS), Vol. 15, No. 5, May 2017
3.2 FAST (Features from Accelerated Segment Test):- It is based on the intensity to extract corner or minimizing the processing data, it is used in image matching. It is to find the interest point in an image, image has pixel which has a well defined position and can be robustly detected .Interest point detection has local information content repeatable between different images. Interest point detection has application in image matching, object recognition, tracking etc.
4. Results & Analysis The Experimentation has been done on the MATLAB environment. The Experimentation applied on the real images of EUS infected fish. a)
Neural Network: - Pattern recognition Neural Network has been implemented to find the classification accuracy. The dataset has been divided into two parts a) Training and b) Testing. The training dataset is used to train the set and Testing is to test the dataset and find the accuracy with the True positive and Negative.
4.1 Confusion Matrix after Applying the Neural Network
Figure 3a) HOG-PCA-NN
Figure 3 b) FAST-PCA-NN
After applying the Neural Network algorithm find the classification accuracy through the confusion Matrix, the result shows in Figure 3 a & b FAST-PCA-NN gives better accuracy as compared to the HOG-PCA-NN. The confusion Matrix shows the True positive & Negative and False positive & Negative in all cases and classify the infected and non-infected fish. The FAST-PCA-NN gives 86% accuracy and HOG-PCA-NN gives 65.8%
219
https://sites.google.com/site/ijcsis/ ISSN 1947-5500
International Journal of Computer Science and Information Security (IJCSIS), Vol. 15, No. 5, May 2017
4.2 Comparison between FAST-PCA-NN & HOG-PCA-NN
Figure 4 Comparisons between HOG-PCA-NN & FAST-PCA-NN
It shows the Performance comparison between the two feature extractor after applying the same classifier and result shows that FAST-PCA-NN gives better accuracy as compared to the HOG-PCA-NN.
5. Conclusion The feature Descriptors shows good result after applying the PCA as it reduce the dimensionality reduction and gives better result .FAST-PCA-NN gives better result as compare to the HOG-PCA-NN. It gives 20.2% better accuracy. Feature Descriptors extract the features and give the output. Neural Network has been applied for classification accuracy as it is an efficient technique and it gives better result. The Experimentation has been done on MATLAB Environment on real images of EUS Infected fish.
Acknowledgements The EUS infected images have been collected from National Bureau of Fish Genetic Resources (NBFGR, Lucknow) and ICAR-Central Inland Fisheries Research Institute (CIFRI), Kolkata. Thanks to Dr. A.K Sahoo(CIFRI,Kolkata) and Dr .P.K Pradhan(NBFGR,Lucknow).
REFERENCES
[1].
Navneet Dalal, Bill Triggs,”Histograms of Oriented Gradients for Human Detection”International Conference on Computer Vision & Pattern Recognition (CVPR ’05), Jun 2005, San Diego, United States.
[2].
Valentin Lyubchenko,Rami Matarneh,Oleg Kobylin,Vyacheslav Lyashenko, Digital Image Processing Techniques for Detection and Diagnosis of Fish Diseases, International Journal of Advanced Research in Computer Science and Software Engineering, Volume 6, Issue 7, July 2016
220
https://sites.google.com/site/ijcsis/ ISSN 1947-5500
International Journal of Computer Science and Information Security (IJCSIS), Vol. 15, No. 5, May 2017
[3].
A. Jeyanthi Suresh , P.Asha ,Human Action Recognition in Video using Histogram of Oriented Gradient (HOG) Features and Probabilistic Neural Network (PNN), International Journal of Innovative Research in Computer and Communication Engineering, Vol. 4, Issue 7, July 2016
[4].
E. Rosten, R. Porter, T. Drummond FASTER and better: a machine learning approach to corner detection IEEE Trans Pattern Anal Mach Intel, 32 (2010), pp. 105–119
[5].
Abdi. H., & Williams, L.J. (2010). "Principal component analysis". Wiley Interdisciplinary Reviews: Computational Statistics. 2 (4): 433–459. doi:10.1002/wics.101.
[6].
Shaw P.J.A. (2003) Multivariate statistics for the Environmental Sciences, Hodder-Arnold. ISBN 0-34080763-6
[7].
Jeong-Seon Park, Myung-Joo Oh, and Soonhee Han, Fish Disease Diagnosis System Based on Image Processing of Pathogens Microscopic Images, Frontiers in the Convergence of Bioscience and Information Technologies 2007
[8].
Burge, C. A., Mark Eakin, C., Friedman, C. S., Froelich, B., Hershberger, P. K., Hofmann, E. E., Ford, S. E. (2014). Burge, C. A., Mark Eakin, C., Friedman, C. S., Froelich, B., Hershberger, P. K., Hofmann, E. E., Ford, S. E. (2014). Climate change influences on marine infectious diseases: implications for management and society. Annual review of marine science, 6, 249-277.
[9].
Lafferty, K. D., Harvell, C. D., Conrad, J. M., Friedman, C. S., Kent, M. L., Kuris, A. M., Saksida, S. M. (2015). Infectious diseases affect marine fisheries and aquaculture economics. Annual review of marine science, 7, 471-496.
[10]. Afferty, K. D., Harvell, C. D., Conrad, J. M., Friedman, C. S., Kent, M. L., Kuris, A. M., Saksida, S. M. (2015). Infectious diseases affect marine fisheries and aquaculture economics.,Annual review of marine science, 7, 471-496. [11].
Narasimha-Iyer, H., et al., Automatic Identification of Retinal Arteries and Veins From DualWavelength Images Using Structural and Functional Features. Biomedical Engineering, IEEE Transactions on, 2007. 54(8): p. 1427-1435.
[12]. Daoliang Lia, Zetian Fua, Yanqing Duanb,Fish-Expert: a web-based expert system for fish disease diagnosis, Expert Systems with Applications 23 (2002) 311–320. [13]. Nayak, K. K., Pradhan, J., Das, B. K. (2014). Original Research Characterization, pathogenicity, antibiotic sensitivity and immune response of Flavobacterium columnare isolated from Cirrhinus mrigala and Carassius auratus. Int. J. Curr. Microbiol. App. Sci, 3(11), 273-287. [14]. J.-Y.Chang and J.-L. Chen, “Automated facial expression recognition system using neural networks”, Journal of the Chinese Institute of Engineers, vol. 24, no. 3, (2001), pp. 345-356. [15]. M. Rizon, M. F.Hashim, P. Saad, S. Yaacob, “Face Recognition using Eigenfaces and Neural Networks”, American Journal of Applied Sciences, vol. 2, no. 6, (2006), pp. 1872-1875.
221
https://sites.google.com/site/ijcsis/ ISSN 1947-5500
