SECURE IMAGE RETRIEVAL BASED ON HYBRID FEATURES AND HASHES
Download as PPT, PDF2 likes620 views
This document describes a system for secure image retrieval based on hybrid features and hashes. It discusses using SURF features to extract information from images and applying a two column histogram hashing algorithm to generate hash codes for images. These hash codes are stored in a database with 1000 images. When a query image is input, its features are extracted and hashed to find similar images in the database based on hamming distance of hash codes. The system is able to retrieve similar and exact match images with precision ranging from 0.46 to 0.93 on sample image topics like buses, dinosaurs and chocolate based on evaluations. Future work could involve improving accuracy of the system for larger databases.
![REVIEW OF LITERATURE SURVEY
Sr. No Methodology Description
1. Color coherence vector and Sobel
Edge Detection[1,2]
1. Exacts contents from that image using Sobel Edge detection .
2. Stores its content information in a coherence vector is used for
matching two images.
3. If difference between query image and image stored in DB is found
to be greater than 1000 then the difference is considered to be
significant and the change is detected.
4. However many a times this technique alone fails to find similarity
between two images. Hence is not so effective.](https://image.slidesharecdn.com/defencepptranjit-150820161525-lva1-app6892/85/SECURE-IMAGE-RETRIEVAL-BASED-ON-HYBRID-FEATURES-AND-HASHES-6-320.jpg)
![Sr. No Methodology Description
2. New Wavelet Features [3,4] 1. It selects sixty four largest Haar wavelet coefficients in each of the
3 color band and stores them in feature vector as +1 or -1 along with
their position in the transformation matrix.
2. Low frequency coefficients tend to be more dominant than those
of the high frequency coefficients and this makes this algorithm
ineffective for images with sharp color changes.
3. Haar wavelet basis is not suitable for natural images.
3. Image Feature Inverted Indexing
[5,7]
1. it extracts image shape feature by using Zernike
moments.
2. Eva extracts image texture feature by using Gray
level co-occurrence matrix algorithm depicting characteristic of
texture.
3. It retrieves seven texture characteristics: uniform, contrast,
correlation, variance, inverted-deviation-matrix, standard-average,
standard deviation.
4. Content Oriented Image Retrieval
(COIR)
[14][15]
1.Divides image into several regions based on visual characteristics
considering as object.
2. The indexing engine calculates the visual attribute values for each
region.
3. The extracted attribute values are stored as metadata and used
during the query processing phase to determine similarity.](https://image.slidesharecdn.com/defencepptranjit-150820161525-lva1-app6892/85/SECURE-IMAGE-RETRIEVAL-BASED-ON-HYBRID-FEATURES-AND-HASHES-7-320.jpg)
![Two column histogram hashing algorithm
While applying TCH [3][11] algorithm first divide that image into number of cell
i.e. NH × NH
After that there is calculation of mean and variance of pixels in each cell region
(m=0,1,……, NH; n=0,1,…..,Nh) and taking this as projection formula
parameters, then there can compare the gray difference of each pixel with certain
combinations of the mean and the variance and then finally, transform the pixel
in each cell into a ternary value.
Input image
Image divided into
no. of cells](https://image.slidesharecdn.com/defencepptranjit-150820161525-lva1-app6892/85/SECURE-IMAGE-RETRIEVAL-BASED-ON-HYBRID-FEATURES-AND-HASHES-13-320.jpg)
SECURE IMAGE RETRIEVAL BASED ON HYBRID FEATURES AND HASHES
- 1. 1 SECURE IMAGE RETRIEVAL BASED ON HYBRID FEATURES AND HASHES 1
- 2. INDEX 2 •Introduction •Problem Definition •Review of Literature survey •Tools and platform •Flow of work •Methodology •Design/ implementation •Results and discussion •Conclusion •Future scope •Publication details •References
- 3. INTRODUCTION Scalable image search based on visual similarity has been an active topic of research in recent year. Crowd of images Facebook Twitter Overloading of similar images. Personal collection of images. Database organization needs proper arrangement. State-of-the-art solution often use hashing method to embed high-dimension image features into Hamming space, where search can be performed on Hamming distance of compact hash codes. The main aim of Image Search using hashing is to search data either related to the query image or the images related to that image or exact images by input as image. The searching process of these search system comprises of first indexing the images on the base of hashing then arranges them in proper manner and then perform matching of the query image with the images and retrieve the desired output images. 3
- 4. PROBLEM DEFINITION Early techniques used in image search engines were not based on visual features but on the textual annotation of images. In those techniques, images were first annotated with text and then searched using a text-based approach from traditional database management systems. However, since automatically generating descriptive texts for a wide range of images were not practical, most text-based image retrieval systems require manual annotation of images. Marking images manually is a bulky and expensive task for large image databases, and is often subjective, context- sensitive and incomplete. As a result, it is difficult for the traditional text-based methods to support a variety of task-dependent queries. 4
- 5. AIM OF RESEARCH • Main objective of Image Search system is to overcome all the difficulties of Textual Annotation. • However, a lot of times, there is a need for searching with a given image and finding related images and also to find exact match of that image i.e. instead of giving a text as an input to search, an image is given as an input. • This system will not only be focusing or searching the information related to the query image. It will also be searching either the similar images or the exact match images. • Eg.- Input as image Image Database
- 6. REVIEW OF LITERATURE SURVEY Sr. No Methodology Description 1. Color coherence vector and Sobel Edge Detection[1,2] 1. Exacts contents from that image using Sobel Edge detection . 2. Stores its content information in a coherence vector is used for matching two images. 3. If difference between query image and image stored in DB is found to be greater than 1000 then the difference is considered to be significant and the change is detected. 4. However many a times this technique alone fails to find similarity between two images. Hence is not so effective.
- 7. Sr. No Methodology Description 2. New Wavelet Features [3,4] 1. It selects sixty four largest Haar wavelet coefficients in each of the 3 color band and stores them in feature vector as +1 or -1 along with their position in the transformation matrix. 2. Low frequency coefficients tend to be more dominant than those of the high frequency coefficients and this makes this algorithm ineffective for images with sharp color changes. 3. Haar wavelet basis is not suitable for natural images. 3. Image Feature Inverted Indexing [5,7] 1. it extracts image shape feature by using Zernike moments. 2. Eva extracts image texture feature by using Gray level co-occurrence matrix algorithm depicting characteristic of texture. 3. It retrieves seven texture characteristics: uniform, contrast, correlation, variance, inverted-deviation-matrix, standard-average, standard deviation. 4. Content Oriented Image Retrieval (COIR) [14][15] 1.Divides image into several regions based on visual characteristics considering as object. 2. The indexing engine calculates the visual attribute values for each region. 3. The extracted attribute values are stored as metadata and used during the query processing phase to determine similarity.
- 8. TOOLS AND PLATFORM • Framework – – Technology: ASP.NET 4.5 – Language: C#.NET – Database:SQL SERVER 2008 – Operating system: WINDOWS 7 • Hardware - Minimum 4GB RAM, 500GB Hard Disk
- 9. FLOW OF WORK Fig. 1. Flow of Work
- 10. METHODOLOGY Following is the step-by-step process of how the system works: 1)Here there is maintaining one database by adding images into it after extracting features of image by SURF algorithm and then apply hashing technique on it, in this manner near about a database of 1000 image is generated as shown below, with hash code in database. 2)This database contains different-different image set like images of bus, animals, nature, etc. SURF (Speed Up Robust Features) SURF (Speed Up Robust Features) algorithm, is base on multi-scale space theory and the feature detector is base on Hessian matrix. Since Hessian matrix has good performance and accuracy. In image I, x = (x, y) is the given point, the Hessian matrix H(x, σ) in x at scale σ, it can be define as
- 11. Where Lxx (x, σ) is that the convolution results of the second order derivative of Gaussian filter with the image I in purpose x and equally for Lxy (x, σ) and Lyy (x, σ), SURF creates a “stack” whereas not 2:1 down sampling for higher levels at intervals the pyramid resulting in photos of an identical resolution. Thus the use of integral photos, SURF filters the stack using a box filter approximation of second-order Gaussian partial derivatives. Since integral photos allow the computation of rectangular box filters in near constant time. Algorithm Detected feature Points Matching feature point Feature matching Time Image1 Image2 SIFT (Scale Invariant Feature Transform) 892 934 41 1.543 s SURF (Speed Up Robust Feature) 281 245 28 0.546 s Comparison of Surf & Shift
- 12. 3) First while adding image into database, extract the features of that image and then on the base of that extracted feature calculate the hash code of image which is run through two column histogram hashing algorithm which is work as explain above. 4) Similar manner while searching images from database there is use one image as an input query, that image is again then go through the same procedure and on the base of that extracted feature there is generate hash code which is unique for that image and then that hash code is search into database or similar images of that type or near neighbor is search into database and if similar code is found into database then it fetch that image and display into screen. 5) Here one example of it that a bus take as an query image and then search those image into database that image first go through feature extraction process then on the base of that extracted feature hash code generate of that image which go through two column histogram hashing algorithm then that hash code is search into database. As shown below,
- 13. Two column histogram hashing algorithm While applying TCH [3][11] algorithm first divide that image into number of cell i.e. NH × NH After that there is calculation of mean and variance of pixels in each cell region (m=0,1,……, NH; n=0,1,…..,Nh) and taking this as projection formula parameters, then there can compare the gray difference of each pixel with certain combinations of the mean and the variance and then finally, transform the pixel in each cell into a ternary value. Input image Image divided into no. of cells
- 14. 1000 chosen images along with the query image are processed using Hashing to compute a 64 bit hash code for each single image. Here Hashing uses SURF and TCH it takes care of the variations like rotation, contrast, brightness, aspect ratio etc. and is thus able to generate good results, then calculate the Hamming Distance between hash code of the query image and hash code of the 1000 images and pick all the images as our final result, which falls under a desired Hamming radius.
- 15. There is maintaining a database of number of different-different images. The system has 1000 images of different resolutions. Then indexing of those images was done in the database. Fig. 2. Home Screen of Reverse Image Search Engine User need to upload the query image for which it wants to perform searching and once the user clicks on search the system needs to follow some steps. DESIGN/ IMPLEMENTATION
- 16. Fig. 3. Uploaded Query Image Then process all necessary steps that is first extract the features of image and on the base of that apply TCH hashing algorithm on it. In this divide image into number of cells and then calculate the mean and variance of pixels in each cell region, after that compare the gray difference of each pixels with certain combination of the mean and variance and then finally, transform the pixels in each cell into a ternary value.
- 17. Fig. 5. Similar images retrieved to the query image Fig. 4. After Feature Extraction and Implementation TCH
- 18. RESULTS AND DISCUSSION Table 1. Showing precision vs. recall of topics Image No. of images in database No. of retrieve images No. of relevant image Precision Recall Bus 32 16 14 0.87 0.43 32 16 15 0.93 0.46 Dinosaur 57 28 14 0.50 0.24 57 28 15 0.53 0.26 chocolate 120 60 30 0.50 0.25 120 60 28 0.46 0.23
- 19. Fig. 6. Graph showing precision vs. recall of topics
- 20. CONCLUSION • A regular image search on any search engine begins by entering a keyword, which is followed by a set of images related to that image. However, a lot of times, there is a need for searching with a given image and finding related images and also to find exact match of that image • It performs indexing and storing of images in DB and thus makes this Search engine suitable for large image databases. Also it is capable of retrieving of similar and exact images this makes the image Search efficient for Large Image Databases. 20
- 21. FUTURE SCOPE • It is capable of searching of both similar and exact images based on Content and Feature but the system is not much capable of retrieving similar images based on the modified versions of the query image. • And since it is not accurate in doing so it cannot be implemented on large image databases which can then be used for detecting unauthorized use of brands and copyright images. • The system now contains only few thousand of images hence the size of the system need to be expanded. This Search only identifies the images undergone different features but doesn’t identify images that have cropped or have some illumination changes.
- 22. PUBLICATION DETAILS Sr.N o Authors Title of the paper Journal Name Volume Issue ISSN Impact Factor Conference Name Venue Date Publisher Page No. 1. Ranjit Banshpal, N.V.Mahajan Review on Different Image Retrieval Techniques International Journal of Advanced Research in Computer Science and Software Engineering. Volume 4, Issue 11, ISSN: 2277 128X. Impact facter-2.080 279-281 2. Ranjit Banshpal, N.V.Mahajan Efficient Image Search using Hybrid Features and TCH International Conference on Pervasive Computing ICPC 2015, Pune, 8-9/1/2015, IEEE, ISBN- Proceedings : 978- 1-4799-6272-3/14/ ---
- 23. REFERENCES 23 1. Divya venkata, divakar yadav.” Image query based search engine using content based image retrieval”. 2012 14th International Conference on Modelling and Simulation. 2. Divya Ragadha, Deepika Kulshreshtha and Divakar Yadav ”Techniques for refreshing Images on Web Document ”, 2011 International Conference on Control, robotics and Cybernetics. 3. A. Lakshmi Subrata Rakshit “New Wavelet features for image indexing and retrieval”. 2010 IEEE 2nd International Advance Computing Conference. 4. Praveen Kumar, Aparna, Dr K Venkata Rao PhD.” Compact Descriptors For Accurate Image Indexing And Retrieval: Fcth And Cedd”. International Journal of Engineering Research & Technology (IJERT) Vol. 1 Issue 8, October – 2012 ISSN: 2278-0181 5. Tan Hao, Chen Yu, Qiu Hang ,”Implementation Of FII-based image retrieval engine”, Apperceiving Computing and Intelligence Analysis, 2009. ICACIA 2009. International Conference on. 6. Piyush Kansal and Vinay Krishnamurthy ,”Reverse Image Search” . 7. Divya venkata, divakar yadav.” Image query based search engine using content based image retrieval”. 2012 14th International Conference on Modelling and Simulation. 8. Chao Zhou, Yangxi Li, Bo Geng, Chao Xu. “Learning a scalable ranking model for content based image retrieval”.IEEE Trans on Image Processing Vol 20, N0-9,Sept 2012.
- 24. 24 9. Mussarat Yasmin , Sajjad Mohsin.”Image Retrieval by Shape and Color Contents and Relevance Feedback”.2012 10th International Conference on Frontiers of Information Technology. 10. Neeraj Kumar, Alexander C. Berg, Peter N. Belhumeur, and Shree K. Nayar.” Describable Visual Attributes for Face Verification and Image Search”. IEEE transaction on pattern analysis and machine intelligence, vol.33, no-10,October 2011. 11. P M Panchal, S R Panchal and S K Shah, “A Comparison of SIFT and SURF”, International Journal of Innovative Research in Computer and Communication Engineering Vol. 1, Issue 2, April 2013. 12. Yu-Gng Jiang, Jun Wang, Member, IEEE, Xiangyang Xue, Member, IEEE, and Shih-Fu Chang, Fellow, IEEE, "Query-Adaptive Image Search with Hash Codes”, IEEE TRANSACTIONS ON MULTIMEDIA, VOL. 15, NO. 2, FEBRUARY 2013. 13. Bin Li, Delie Ming, Wenwen Yan, Xiao Sun, Tian Tian, and Jinwen Tian, “Image Matching Based on Two-Column Histogram Hashing and Improved RANSAC”, IEEE GEOSCIENCE AND REMOTE SENSING LETTERS, VOL. 11, NO. 8, AUGUST 2014. 14. Tamanna saini, “One Time Password Generator System”, International Journal of Advanced Research in computer Science and Software Engineering Vol. 4, Issue 3, April 2014. 15. A. W. Smeulders, M.Worring, S. Santini, A. Gupta, and R. Jain, “Content-based image retrieval,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 22, no. 12, pp. 1349–1380, Dec. 2000.
- 25. 16. Soo-Chang Pei, Senior Member, IEEE, and Ching-Min Cheng." Extracting Color Features and Dynamic Matching for Image Data-Base Retrieval". IEEE Transactions On circuits and systems for video technology, VOL. 9, NO. 3, APRIL 1999. 17. Xinmei Tian, Dacheng Tao, Member, IEEE, Xian-Sheng Hua, Member, IEEE, and Xiuqing Wu."Active Re-ranking for Web Image Search". IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 19, NO. 3, MARCH 2010. 18. Ranjit Banshpal, Nikita Mahajan. "Efficient Image Search using Hybrid Features and TCH”, International Conference on Pervasive Computing ICPC 2015, Pune, IEEE, ISBN-Proceedings: 978-1-4799-6272-3/14 19. Ranjit Banshpal, Nikita Mahajan. "Review on Different Image Retrieval Techniques", International Journal of Advanced Research in Computer Science and Software Engineering, Volume 4, Issue 11, ISSN: 2277 128X.
- 26. 26 THANK YOU