SHORT PAPER International Journal of Recent Trends in Engineering, Vol 2, No. 3, November 2009
Image Comparison Search Engine Based On Traditional and Improved Fractal Encoding Techniques. Shraddha Viraj Pandit 1 , M.V. Kulkarni 2, M.L.Dhore 3. 1 M.E. (CSE-IT) Student Department of Computer Engg.,Vishwakarma Institute of Technology,University of Pune, Pune,India
[email protected], 2 Assistant Professor Department of Computer Engg,Vishwakarma Institute of Technology,University of Pune, Pune, India
[email protected], 3 Associate Professor and head of Department of Computer Engg, Vishwakarma Institute of Technology, University of Pune,Pune,India
[email protected] 1.
The ICSE makes use of the comparison mode and returns correct or similar images from the database and processes the query without knowing the image filename. It is the image Eigen value database as a search engine kernel. 2. The ICSE try to reduce data space by only retrieving the Eigen value of the image by applying fractal image processing of the image in the spatial domain, and store the results in the image Eigen value database [2]. The entire ICSE process is broadly divided into four parts: 1. Image normalization 2. Retrieval of Eigen value from Fractal Image Processing 3. Image storage 4. Analysis (Eigen value comparison). The image Search Engine Work Flow Chart is shown in Fig.1. Input Original Image
Abstract— This search engine allows users to quickly obtain information from networks. Traditional search engines can only search the data of modal characters. To solve this problem, Image Comparison Search Engine (ICSE) makes use of “Fractal Image processing “to create a database using image Eigen values. When a user input is an image query, this system will generate image Eigen value data, compare this with the data in the database of image Eigen value, and output the results. ICSE can not only find the exact input image for the source image, but also find the “right image” when the source image is rotated. Index Terms— fractal image compression, search engine, mean, range block, domain block
I. INTRODUCTION Image search (or image search engine) is a type of search engine specialized on finding pictures, images, animations etc. With the rapid pace of computer technology over the past several years, the information that users use is no longer mainly character based. Traditional character based search engines are unable to provide the capabilities needed for searching image data. A search engine allows users to quickly obtain information from networks. Traditional search engines can only search the data of modal characters. The solution to this problem is to implement an Image Comparison Search Engine (ICSE), and make the use of “Fractal Image Processing” to create a database using image Eigen values. When user input an image query, this system will generate image Eigen value data, compare this with the data in the database of image Eigen value, and output the results. ICSE can not only find the exact input image for the source image, but also find the “right image” when the source image is rotated [2].
Image Normalization Retrieve Eigen Value
Image Storage
Eigen value Comparison
II. ICSE METHODOLOGY
Fig.1. Image Search Engine Work Flow
Here is the effort to design and implement a new search engine, called Image Comparison Search Engine (ICSE), to solve this problem.
A. Image Normalization There are varieties of images on the Internet. Firstly there is a need to normalize the image properties of size, color 108
© 2009 ACADEMY PUBLISHER
SHORT PAPER International Journal of Recent Trends in Engineering, Vol 2, No. 3, November 2009 and intensity [2]. Normalization is a process that changes the range of pixel intensity value. It is a linear process. Image normalization means, if the intensity range of the image is 50 to 180 and the desired range is 0 to 255, then subtracting 50 from each pixel intensity making the range 0 to 130.Each pixel intensity is multiplied by 255/130, making the range 0 to 255. To select the size, in order to reduce the download time, an image cannot be big, and the average size is approximately 100 × 100 pixels. When a user inputs an image and compares it with other images in the database, through intensity normalization it results into more similar and accurate images.
such maps, constitute the encoding. The process also does not store the actual pixel values present in the domain, instead stores the index values notifying about the domain block, reducing the further storage required [1]. III. IMPROVED FRACTAL ENCODING ALGORITHM The Improved Fractal Encoding Algorithm is different from the Traditional (Fixed Point) Fractal Encoding Algorithm. Under the improvement of fractal gain there is a need to compare a range block with a domain block through RMS method. The Fig.2 shows the Eigen Value retrieving flow. Divide images into 4 × 4 range blocks and 8 × 8 domain blocks. Then each range block is compared with the domain blocks. When similar, an 8 × 8 domain block and the corresponding four 4 × 4 range blocks are marked with a flag. This indicates that an 8 × 8 domain block is substituted by a range block, and also the related four 4 × 4 range blocks with in the 8 × 8 domain block need not be compared with other domain blocks. The position, size, corresponding range block position, and range block information of the marked domain block are stored in the hash table. When all 8 × 8 domain blocks are being checked by a range block, then separate these domain blocks but similar range blocks, and divide each domain block into four 4 ×4 domain blocks. Then they are compared with the range blocks [2].
B. Retrieval of Eigen Value Image Comparison Search Engine is the Eigen Value database as a search engine kernel. Eigen values are the multipliers. In the ICSE system Eigen values are used for comparison purpose. In a traditional character type search engine, characters take less space. One single diskette can store a great many words. However, since images take a large amount of storage, it is impractical to store a whole image in the database. Therefore, in this system, we try to reduce data space by only retrieving the Eigen values of the image by applying fractal image processing of the image in the spatial domain, and store the results in the image Eigen value database [2].For comparing the images two types of filters are used: 1. Mean The mean is a basic and fast method of image processing. Using the mean as the first step of screening may eliminate images with large color difference. It sums all the pixel image color values and then divides by the number of pixels.
IV. IMAGE STORAGE Database gives facility to store small images and data. Instead of storing whole image in the database store link into the database. After all domain blocks have been compared with range blocks, to obtain the Eigen value from the hash table and store it into the database. In this relational database, all information is divided into 2 tables.ImageDetails, ImageData. In ImageDetails table, there are five entities: ImageId, ImageName, ImageWidth, ImageHeight and Mean. In the ImageData table, there are two entities: ImageId and Eigen value [2].
2. Fractal Image Compression Technique A fractal is nothing but the rough fragmented geometric shape that can be subdivided in parts, each of which is approximately a reduced size copy of the whole. Hence the fractals can be subdivided into number of parts such that each small scale part looks similar to the whole object [1] [4]. 3. Fractal Encoding Algorithm The process of encoding with fixed partitioning (Traditional Encoding Scheme) is explained here. According to PIFS (Partitioning Iterated Function System), the process of encoding emphasizes to divide the image into two categories of square blocks: Range blocks and Domain blocks [1]. The domains are twice the range size; therefore the pixels in the domain are averaged in groups of four so that the domain is reduced to the size of the range. In the process of encoding the range blocks are compared with all the potential domains from the domain pool by applying the affine transformations on the pixel values. The target of this process is to find the best matching pair of range and domain blocks that minimizes the root mean square error value based on the scaling si, and the offset oi. When such a pair is found, the optimal domain block and the affine transformations on the pixel values are stored - this constitutes one map wi. The collection of all
. Fig.2. Retrieve Eigen value Flow Chart [2].
V. ANALYSIS 109
© 2009 ACADEMY PUBLISHER
SHORT PAPER International Journal of Recent Trends in Engineering, Vol 2, No. 3, November 2009 For an image search, it stores all Eigen values from fractal image processing and the mean into the database. In comparison with a data, when user inputs an image, the system will normalize it, calculate the mean and get the Eigen value from fractal image processing, and record the Eigen value into the Hash table. Image search engine sets up two filters when making a comparison. The first filter uses mean and second filter uses Eigen value for image comparison [2]. For experimentation we have used a Pentium IV, 512MB RAM computer. The operating system is WindowsXP. Programming is done in Java1.6.0.In these experimentation, Photoshop is used to make the modifications to the images. The images used in the project are the grayscale images. Total 50 images were downloaded from the Internet. Out of these 35 images are used for query purpose. Each image goes through the required three steps. The following tables show how much time (time in milliseconds) was spent matching the given images using Traditional Fractal Encoding Scheme (TFE) and Improved Fractal Encoding scheme (IFE) at different RMS values. By observing all the tables we can say that RMS value 90 is the best RMS for comparing the different images.
Fig. 4. Plot for RMS values vss.Time for image-boat.
The Fig.5 shows output of the searched image lenna64 by Traditional Fractal Encoding Scheme and Improved Fractal Encoding Scheme. The standard Lenna image has been taken as a original image (searched image) which is shown in Fig.5 (a) .The Fig. 5 (b) shows the Normalized image. The Fig.5 (c) shows the output of the searched image by Mean method. The Fig.5 (d) shows the output of the searched image by Eigen value method
Table No.1: Execution Speed of Fixed Point Encoding and Improved Fractal Encoding at RMS: 110. Sr.No. Image Size Traditional Improved Fractal Fractal Encoding Encoding (Time (Time in in ms) ms) 1 Lena64 64x64 18937 18078 2 Barbara 64x64 16750 16391 copy 3 Boat64 64x64 19094 17531
Fig. 5(a). Searched Image
Fig. 5(b). Normalized Image
Fig. 5(c). Images found by Mean method Table No.2: Execution Speed of Fixed Point Encoding and Improved Fractal Encoding at RMS: 90. Sr.No. Image Size Traditional Improved Fractal Fractal Encoding Encoding (Time in (Time in ms) ms) 1 Lena64 64x64 16578 16219 2 Barbara 64x64 16687 16672 copy 3 Boat64 64x64 17781 17500
Fig. 5 (d). Image found by Eigen value method
CONCLUSIONS By looking in detail at all the features of Image Comparison Search Engine and the entire process of Fractal Image Compression, we can conclude that Image Compression Search Engine can find the exact input image for the source image, but also find the right image when the source image is rotated.
The different graphs indicate the variations in time taken for searching the images at different RMS values.
REFERENCES [1] Fisher, Y., “Fractal Image Compression Theory and Application, Springer-Verlag, New York, 1994. [2] Kwang-Fu Li and Tung-Shou Chen,”Fractal Image Process Based Image Comparison Search Engine“, Taiwan, 2003. [3] C.J.Cheng and R.F.Chang,” Fractal coding for Still images”, Master Thesis, CCU, Taiwan, R.O.C, 1996. [4] M.Ali and T.G.Clarkson,”Fractal image Compression”, ITA91. [5] Rafale C.Gonzalez and Richard E.Woods,”Digital Image Processing”, second edition.
Fig.3. Plot for RMS values vss.Time for image-lenna64.
110 © 2009 ACADEMY PUBLISHER
