VIEWS: 104 PAGES: 6 CATEGORY: Emerging Technologies POSTED ON: 9/5/2010
IJCSIS invites authors to submit their original and unpublished work that communicates current research on information assurance and security regarding both the theoretical and methodological aspects, as well as various applications in solving real world information security problems.
(IJCSIS) International Journal of Computer Science and Information Security, Vol. 08, No.05, 2010 Image Retrieval using Row Means of Column Transformed Even and Odd Parts of Image with Walsh, Haar and Kekre Transforms Dr. H.B.Kekre1, Sudeep D. Thepade2, Akshay Maloo3 1 Senior Professor, 2Ph.D.Research Scholar & Associate Professor, 3B.Tech (CE) Student Computer Engineering Department, MPSTME, SVKM‟s NMIMS (Deemed-to-be University), Mumbai, India 1 email@example.com, firstname.lastname@example.org,email@example.com Abstract— The paper presents an innovative content based image [5,18], criminal investigations , image search on the retrieval (CBIR) technique based on row mean of column Internet [9,19,20]. transformed even and odd part of image. Here the performance of proposed CBIR technique is tested for different transforms A. Content Based Image Retrieval like Walsh transform, Haar transform and Kekre transform. In literature the term content based image retrieval (CBIR) Instead of using all pixels of image as feature vector for image has been used for the first time by Kato et.al. , to describe retrieval, row mean of transform applied on columns of odd and his experiments into automatic retrieval of images from a even part of image is used, resulting into better performance with database by colour and shape feature. The typical CBIR system much lower computations. The proposed CBIR techniques are tested on generic image database having 1000 images spread performs two major tasks [16,17]. The first being feature across 11. For each transform 55 queries (5 per category) were extraction (FE), where a set of features, called feature vector, is fired. Then the precision and recall for all queries are computed. generated to accurately represent the content of each image in While considering the relevance of result images for calculation the database. The second task is similarity measurement (SM), of precision and recall the results of odd image part and even where distance between the query image and each image in the image part both are ORED. The average precision and average database using their feature vectors is used to retrieve the top recall values of all queries gives the performance comparison of “closest” images [16,17]. For feature extraction in CBIR there proposed CBIR methods as compared to considering full image are mainly two approaches  feature extraction in spatial data as feature vector or considering the simple row mean as domain and feature extraction in transform domain. The feature feature vector. The results have shown the performance extraction in spatial domain includes CBIR techniques based improvement (higher precision and recall values) with proposed on histograms , BTC [1,2,16], VQ . The transform methods compared to all pixel data of image at reduced domain methods are widely used in image compression, as they computations resulting in faster retrieval. The comparison of give high energy compaction in transformed image [17,24]. So transforms for CBIR performance gives Walsh transform it is obvious to use images in transformed domain for feature surpassing the other two considered here. extraction in CBIR . But taking transform of image is time consuming, this complexity is reduced to a great extent by the Keywords-content based image retrival (CBIR), row mean, proposed technique. Reducing the size of feature vector using Walsh transform, Haar transform, Kekre transform. pure image pixel data in spatial domain only and till getting the I. INTRODUCTION improvement in performance of image retrieval is the theme of the work presented. Many current CBIR systems use Euclidean The large numbers of images are being generated from a distance [1-3,8-14] on the extracted feature set as a similarity variety of sources (digital camera, video, scanner, the internet measure. The Direct Euclidian Distance between image P and etc.) which have posed technical challenges to computer query image Q can be given as equation 1, where Vpi and Vqi systems to store/transmit and index/manage image data are the feature vectors of image P and Query image Q effectively to make such large collections easily accessible. respectively with size „n‟. Image compression deals with the challenge of storage and transmission, where significant advancements have been made n [1,4]. The challenge to image indexing is studied in the context ED (Vpi Vqi ) i 1 2 (1) of image database [2,6,7,10,11], which has become one of the most promising and important research area for researchers from a wide range of disciplines like computer vision, image II. IMAGE TRANSFORMS processing and database areas. The thirst of better and faster image retrieval techniques is increasing day by day. Some of The various transforms [5,24] used for proposed CBIR important applications for CBIR technology could be identified techniques are discussed below: as art galleries [12,14], museums, archaeology , architecture design [8,13], geographic information systems , trademark databases [21,22], weather forecast [5,21], medical imaging 79 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 (IJCSIS) International Journal of Computer Science and Information Security, Vol. 08, No.05, 2010 A. Walsh Transform diagonal and diagonal values of Kekre‟s transform matrix are The Walsh transform matrix [5,18] is defined as a set of N one, while the lower diagonal part except the values just below rows, denoted Wj, for j = 0, 1, ... , N - 1, which have the diagonal is zero. following properties: Generalized NxNKekre‟s transform matrix can be given as: • Wj takes on the values +1 and -1. 1 1 1 .. 1 1 • Wj = 1 for all j. N 1 1 1 .. 1 1 • Wj xWkT=0, for j k and Wj xWkT =N, for j=k. 0 N 2 1 .. 1 1 • Wj has exactly j zero crossings, for j = 0, 1, ...., N-1. K NxN (2) • Each row Wj is either even or odd w.r.t. to its midpoint. 0 0 0 .. 1 1 0 0 0 .. N ( N 1) 1 B. Haar Transform The formula for generating the term Kxy of Kekre‟s This sequence was proposed in 1909 by AlfrédHaar. Haar transform matrix is: used these functions to give an example of a countable orthonormal system for the space of square-integrable (3) functions on the real line . The study of wavelets, and even the term "wavelet", did not come until much later. The Haar wavelet is also the simplest possible wavelet. The technical disadvantage of the Haar wavelet is that it is not III. PROPOSED TECHNIQUE continuous, and therefore not differentiable. This property can, however, be an advantage for the analysis of signals with The CBIR technique given in  has been extended here. sudden transitions, such as monitoring of tool failure in In  the feature vector considered for CBIR is row mean of machines. column transformed image. Here first the odd part and even part of image are obtained using it‟s mirror image. Then transform is applied on each column of obtained images to C. Kekre‟s Transform calculate feature vector, image retrieval is done using both Kekre‟s transform matrix is the generic version of Kekre‟s feature vectors (F1 & F2). The obtained results are combined LUV color space matrix [1,18]. Kekre‟s transform matrix can using OR operator to obtain the final results. Figure 1 shows be of any size NxN, which need not have to be in powers of 2 the feature extraction in proposed CBIR technique with row (as is the case with most of other transforms). All upper mean  of transformed image columns. Figure 1 Feature Extraction in Proposed CBIR Technique using row means of column transformed even and odd part of image 80 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 (IJCSIS) International Journal of Computer Science and Information Security, Vol. 08, No.05, 2010 R R N R N R Sorted Image Index for Even Part (P1) R R N N R R Sorted Image Index for Odd Part (P2) R R N R R R Result Image Index after using OR operator Figure 2 How OR operator is used in proposed CBIR technique The feature extraction steps for proposed image retrieval method can be given as follows. 1. Calculate the mirror image across Y-axis of the query image. 2. Using the image and its mirror image to calculate the Even Image (P1) and Odd image (P2). When you add both P1 and P2 you get the original image back. 3. Apply transform T on the column of P1 and P2 images of size NxN (INxN) to get column transformed images of the same size (cINxN) cINxN (column transformed) = [TNxN] [INxN] (4) 4. Calculate row mean of column transformed image to get feature vector of size N (instead of N2). We get two feature vectors for Even and Odd part (F1 & F2). 5. Then Euclidean Distance is applied and the results are sorted in ascending order for both feature vectors. The Figure 3 Sample Images from Generic Image Database obtained results are combined using OR operator as [Image database contains total 1000 images with 11 categories] seen in figure 2, to calculate precision and recall for the proposed technique. C. Precision/Recall To assess the retrieval effectiveness, we have used the The proposed technique is applied for three different transforms to see where the best performance is obtained. The precision and recall as statistical comparison parameters [1,2] results are compared with applying transform on full image and for the proposed CBIR techniques. The standard definitions row mean on image (without column transform). for these two measures are given by following equations. The figure 2 shows how OR operator is used in proposed Number _ of _ relevant _ images _ retrieved CBIR techniques. Here the „R‟ is relevant image retrieved and Pr ecision (4) Total _ number _ of _ images _ retrieved „N‟ is non-relevant image retrieved. These relevancies are then considered to compute precision and recall values for all queries. Number _ of _ relevant _ images _ retrieved Re call (5) Total _ number _ of _ relevent _ images _ in _ database IV. IMPLEMENTATION V. RESULTS AND DISCUSSIONS A. The Platform For testing the performance of each proposed CBIR The implementation of the proposed CBIR techniques is technique, per technique 55 queries (5 from each category) are done in MATLAB 7.0 using a computer with Intel Core 2 Duo fired on the database of 1000 variable size generic images Processor T8100 (2.1GHz) and 2 GB RAM. spread across 11 categories. The query and database image matching is done using Euclidian distance. The average B. Database precision and average recall are computed and are plotted The CBIR techniques are tested on the image database  against number of retrieved images. The crossover point of of 1000 variable size images spread across 11 categories of precision and recall gives important performance measure for human being, animals, natural scenery and manmade things. image retrieval techniques. Higher the crossover point is better Figure 3 shows sample image of generic database. will be the performance of CBIR method. 81 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 (IJCSIS) International Journal of Computer Science and Information Security, Vol. 08, No.05, 2010 Figure 4 Crossover Point of Precision and Recall v/s Number of Retrieved Figure 6 Crossover Point of Precision and Recall v/s Number of Retrieved Images using Walsh Transform Images using Kekre Transform In Figure 4 the crossover points of CBIR using Walsh To compare the performance of the considered image transform applied to the full image as feature vector (referred transforms with each other for proposed CBIR techniques, the as „Full‟), simple row mean as feature vector (referred as crossover points of proposed CBIR methods with individual „RM‟) and the proposed CBIR technique with Walsh transform transforms are shown together in figure 6. Here the best (referred as „TRANSFORM-RM‟ here „WALSH-RM‟) are performance (highest precision-recall crossover point value) is shown. It can be noted from figure that proposed CBIR given by Walsh transform. The performance ranking of technique with Walsh transform gives best performance than transforms could be given as Walsh, Haar and then Kekre the other discussed methods. transform. From figure 7 also it can noted that proposed CBIR technique with all transforms is performing better than CBIR using full transformed image as feature vector or CBIR using row mean of image as feature vector. Figure 5 Crossover Point of Precision and Recall v/s Number of Retrieved Images using Haar Transform The comparison of crossover points of CBIR with Haar transform of full image as feature vector (Full), row mean of image as feature vector (RM) and the proposed CBIR Figure 7 Crossover Point of average Precision and Recall v/s proposed technique with Haar transform (HAAR-RM) are compared in technique for all transforms with Full Image figure 5. Also here the proposed CBIR method with Haar transform outperforms other image retrieval techniques. VI. CONCLUSION Figure 6 gives the crossover points of CBIR using Kekre The new CBIR method using row means of column transform applied to the full image as feature vector (referred transformed odd and even parts of the image with OR operator as „Full‟), simple row mean as feature vector (referred as is proposed. The technique is tested using 55 queries fired on „RM‟) and the proposed CBIR technique with Kekre transform generic image database of 1000 images spread across 11 ( „KEKRE-RM‟) are shown. It can be noted from figure that categories. The average precision and average recall values of proposed CBIR technique with Kekre transform gives best these queries have proved the proposed CBIR technique better performance than the other discussed methods. than taking complete transformed image as feature vector and also bettr than simple row mean of image as feature vector. The 82 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 (IJCSIS) International Journal of Computer Science and Information Security, Vol. 08, No.05, 2010 proposed techniques are implemented using three mage  H.B.Kekre, Sudeep D. Thepade, “Color Traits Transfer to Grayscale transforms like Walsh transform, Haar transform and Kekre Images”, In Proc.of IEEE First International Conference on Emerging Trends in Engg. & Technology, (ICETET-08), G.H.Raisoni COE, transform. The performance ranking of transforms for proposed Nagpur, INDIA. Uploaded on online IEEE Xplore. CBIR method can be given as Walsh followed by Haar and  http://wang.ist.psu.edu/docs/related/Image.orig (Last referred on 23 Sept then Kekre. 2008)  H.B.Kekre, Sudeep D. Thepade, “Using YUV Color Space to Hoist the REFERENCES Performance of Block Truncation Coding for Image Retrieval”, IEEE  H.B.Kekre, Sudeep D. Thepade, “Boosting Block Truncation Coding International Advanced Computing Conference 2009 (IACC‟09), Thapar using Kekre‟s LUV Color Space for Image Retrieval”, WASET University, Patiala, INDIA, 6-7 March 2009. International Journal of Electrical, Computer and System Engineering  H.B.Kekre, Sudeep D. Thepade, Archana Athawale, Anant Shah, (IJECSE), Volume 2, Number 3, pp. 172-180, Summer 2008. Available Prathmesh Verlekar, Suraj Shirke,“Energy Compaction and Image online at http://www.waset.org/ijecse/v2/v2-3-23.pdf Splitting for Image Retrieval using Kekre Transform over Row and  H.B.Kekre, Sudeep D. Thepade, “Image Retrieval using Augmented Column Feature Vectors”, International Journal of Computer Science Block Truncation Coding Techniques”, ACM International Conference and Network Security (IJCSNS),Volume:10, Number 1, January 2010, on Advances in Computing, Communication and Control (ICAC3- (ISSN: 1738-7906) Available at www.IJCSNS.org. 2009), pp. 384-390, 23-24 Jan 2009, Fr. Conceicao Rodrigous College  H.B.Kekre, Sudeep D. Thepade, Archana Athawale, Anant Shah, of Engg., Mumbai. Is uploaded on online ACM portal. Prathmesh Verlekar, Suraj Shirke, “Walsh Transform over Row Mean  H.B.Kekre, Sudeep D. Thepade, “Scaling Invariant Fusion of Image and Column Mean using Image Fragmentation and Energy Compaction Pieces in Panorama Making and Novel Image Blending Technique”, for Image Retrieval”, International Journal on Computer Science and International Journal on Imaging (IJI), www.ceser.res.in/iji.html, Engineering (IJCSE),Volume 2S, Issue1, January 2010, (ISSN: 0975– Volume 1, No. A08, pp. 31-46, Autumn 2008. 3397). Available online at www.enggjournals.com/ijcse.  Hirata K. and Kato T. “Query by visual example – content-based image  H.B.Kekre, Sudeep D. Thepade, “Image Retrieval using Color-Texture retrieval”, In Proc. of Third International Conference on Extending Features Extracted from Walshlet Pyramid”, ICGST International Database Technology, EDBT‟92, 1992, pp 56-71 Journal on Graphics, Vision and Image Processing (GVIP), Volume 10,  H.B.Kekre, Sudeep D. Thepade, Akshay Maloo, “Performance Issue I, Feb.2010, pp.9-18, Available online www.icgst.com/ Comparison of Image Retrieval Using Fractional Coefficients of gvip/Volume10/Issue1/P1150938876.html Transformed Image Using DCT, Walsh, Haar and Kekre‟s Transform”,  H.B.Kekre, Sudeep D. Thepade, “Color Based Image Retrieval using CSC International Journal of Image Processing (IJIP), Vol. 4 , Issue 2, Amendment Block Truncation Coding with YCbCr Color Space”, Computer Science Journals, CSC Press, Available online at International Journal on Imaging (IJI), Volume 2, Number A09, Autumn http://www.cscjournals.org 2009, pp. 2-14. Available online at www.ceser.res.in/iji.html (ISSN:  Minh N. Do, Martin Vetterli, “Wavelet-Based Texture Retrieval Using 0974-0627). Generalized Gaussian Density and Kullback-Leibler Distance”, IEEE  H.B.Kekre, Tanuja Sarode, Sudeep D. Thepade, “Color-Texture Feature Transactions On Image Processing, Volume 11, Number 2, pp.146-158, based Image Retrieval using DCT applied on Kekre‟s Median February 2002. Codebook”, International Journal on Imaging (IJI), Volume 2, Number  B.G.Prasad, K.K. Biswas, and S. K. Gupta, “Region –based image A09, Autumn 2009,pp. 55-65. Available online at retrieval using integrated color, shape, and location index”, International www.ceser.res.in/iji.html (ISSN: 0974-0627). Journal on Computer Vision and Image Understanding Special Issue:  H.B.Kekre, Sudeep D. Thepade, “Improving the Performance of Image Colour for Image Indexing and Retrieval, Volume 94, Issues 1-3, April- Retrieval using Partial Coefficients of Transformed Image”, June 2004, pp.193-233. International Journal of Information Retrieval, Serials Publications,  H.B.Kekre, Sudeep D. Thepade, “Creating the Color Panoramic View Volume 2, Issue 1, 2009, pp. 72-79 (ISSN: 0974-6285) using Medley of Grayscale and Color Partial Images ”, WASET  H.B.Kekre, Sudeep D. Thepade, Akshay Maloo “Query by Image International Journal of Electrical, Computer and System Engineering Content Using Colour Averaging Techniques”, Engineering journals, (IJECSE), Volume 2, No. 3, Summer 2008. Available online at International Journal of Engineering, Science and Technology (IJEST), www.waset.org/ijecse/v2/v2-3-26.pdf. Volume 2, Issue 6, Jun.2010, pp.1612-1622, Available online  Stian Edvardsen, “Classification of Images using color, CBIR Distance http://www.ijest.info/docs/IJEST10-02-06-14.pdf. Measures and Genetic Programming”, Ph.D. Thesis, Master of science  H.B.Kekre, Sudeep D. Thepade, Akshay Maloo, “Performance in Informatics, Norwegian university of science and Technology, Comparision of Image Retrieval using Row Mean of Transformed Department of computer and Information science, June 2006. Column Image ”, International Journal on Computer Science and  H.B.Kekre, Tanuja Sarode, Sudeep D. Thepade, “DCT Applied to Row Engineering (IJCSE), Volume 2, Issue 5,2010 Mean and Column Vectors in Fingerprint Identification”, In Proceedings of International Conference on Computer Networks and Security AUTHORS PROFILE (ICCNS), 27-28 Sept. 2008, VIT, Pune.  H.B.Kekre, Sudeep D. Thepade, Akshay Maloo “Performance Dr. H. B. Kekre has received B.E. (Hons.) in Telecomm. Comparison for Face Recognition using PCA, DCT &WalshTransform Engineering. from Jabalpur University in 1958, M.Tech of Row Mean and Column Mean”, Computer Science Journals, (Industrial Electronics) from IIT Bombay in 1960, International Journal oj Image Processing (IJIP), Volume 4, Issue II, M.S.Engg. (Electrical Engg.) from University of Ottawa in May.2010, pp.142-155, available online at http://www.cscjournals.org/ 1965 and Ph.D. (System Identification) from IIT Bombay csc/manuscript/Journals/IJIP/volume4/Issue2/IJIP-165.pdf.. in 1970 He has worked as Faculty of Electrical Engg. and  H.B.kekre, Sudeep D. Thepade, “Improving „Color to Gray and Back‟ then HOD Computer Science and Engg. at IIT Bombay. For using Kekre‟s LUV Color Space”, IEEE International Advanced 13 years he was working as a professor and head in the Computing Conference 2009 (IACC‟09), Thapar University, Patiala, Department of Computer Engg. at Thadomal Shahani INDIA, 6-7 March 2009. Is uploaded and available online at IEEE Engineering. College, Mumbai. Now he is Senior Professor Xplore. at MPSTME, SVKM‟s NMIMS University. He has guided  H.B.Kekre, Sudeep D. Thepade, “Image Blending in Vista Creation 17 Ph.Ds, more than 100 M.E./M.Tech and several using Kekre's LUV Color Space”, SPIT-IEEE Colloquium and B.E./B.Tech projects. His areas of interest are Digital Signal International Conference, Sardar Patel Institute of Technology, Andheri, processing, Image Processing and Computer Networking. He Mumbai, 04-05 Feb 2008. has more than 320 papers in National / International Conferences and Journals to his credit. He was Senior 83 http://sites.google.com/site/ijcsis/ ISSN 1947-5500 (IJCSIS) International Journal of Computer Science and Information Security, Vol. 08, No.05, 2010 Member of IEEE. Presently He is Fellow of IETE and Life Member of ISTE Recently ten students working under his guidance have received best paper awards and two have been conferred Ph.D. degree of SVKM‟s NMIMS University. Currently 10 research scholars are pursuing Ph.D. program under his guidance. Sudeep D. Thepade has Received B.E.(Computer) degree from North Maharashtra University with Distinction in 2003. M.E. in Computer Engineering from University of Mumbai in 2008 with Distinction, currently pursuing Ph.D. from SVKM‟s NMIMS, Mumbai. He has about than 07 years of experience in teaching and industry. He was Lecturer in Dept. of Information Technology at Thadomal Shahani Engineering College, Bandra(w), Mumbai for nearly 04 years. Currently working as Associate Professor in Computer Engineering at Mukesh Patel School of Technology Management and Engineering, SVKM‟s NMIMS University, Vile Parle(w), Mumbai, INDIA. He is member of International Association of Engineers (IAENG) and International Association of Computer Science and Information Technology (IACSIT), Singapore. He has been on International Advisory Board of many International Conferences. He is Reviewer for many reputed International Journals. His areas of interest are Image Processing and Computer Networks. He has about 79 papers in National/International Conferences/Journals to his credit with a Best Paper Award at International Conference SSPCCIN-2008, Second Best Paper Award at ThinkQuest- 2009 National Level paper presentation competition for faculty and Best Paper Award at Springer International Conference ICCCT-2010. Akshay Maloo is currently pursuing B.Tech. (CS) from MPSTME, NMIMS University, Mumbai. His areas of interest are Artificial intelligence, Image Processing, Computer Networks and Security. He has 13 research papers in International Conferences/Journals to his credit. 84 http://sites.google.com/site/ijcsis/ ISSN 1947-5500
Pages to are hidden for
"Image Retrieval using Row Means of Column Transformed Even and Odd Parts of Image with Walsh, Haar and Kekre Transforms"Please download to view full document