A Survey Paper on Profile-Based Image Recommender System for Smartphone

DOI : 10.17577/IJERTV3IS100005

Download Full-Text PDF Cite this Publication

Text Only Version

A Survey Paper on Profile-Based Image Recommender System for Smartphone

Ritesh Chaudhari, Pranali Bhaga,t Nikita Mahakalkar

Departent of computer Engineering BapuraoDeshmukh College Of Engineering, Sevagram-Wardha

Akhil Gotmare

Assistant Professor DepartentOf Computer Engineering

BapuraoDeshmukh College Of Engineering, Sevagram-Wardha

Abstract :Imagerecommender systems have become extremely common recent in search engines and are applied in a verity of web based applications. The most popular ones are properly movies, image, news, book, research, articles, search, quintiles social tag and product in general. However, there are also recommender systems for experts, jokes, restaurant, financial service, life insurance, etc. But this recommender system are for web.As user uses search engines for the searching purpose such as google, yahoo etc.This search engines provides output images as per the user intensions.Re-ranking algorithm is to be used to provide output results which are related to query which is entered by user.Moerover there are millions of web pages on web and all these pages are not related to single user search intension.Hence output result is not so much accurate.

In this paper, we propose a profile based image recommender system for import phone.This image recommender system will able user forretrievalofimages from profile based search engines by using smartphone. Each user will have his own profile and semantic signature of each user search will be stored individually.Each user has to log to their own profile with username and password to access web.This improves the performance of search result as it is refering individual profile to know users search intension.

Keywords–Smartphone Image Recommender, Image Recommender, Profile BasedImage Recommender, ImageRe- ranking.


    With increasing gowth of web information,organization and utilization of web has become more complex.The existing image recommender systems are not available for profile based search.Search engines which are used for the search of images useskeyword or textas a queries. These are provided to search engine.According to keywordortext query which have given to search engine related images are shown on the pages of web application.There is no such type of web application which accepts the direct image sas well as text query as a query. In propose system re-ranking of images will be carried out by considering profile ofinvidualuser.This system is capable of accepting text and direct image as an input.queryand according to query image or query text related images will be shown on web pages of an application. As we are entering client images as a query to find exactly similar image related toquery imageimage processingwillbecarried out to show the related images. Once the image processing performed on the query image and related

    images similar to query image are shown on the web pages, according to how many times that images are used by individual user re-ranking of image will be carried out.

    Fig.1.The conventional image re-ranking framework

    For example- Consider a figure 1, if suppose the user is providingapple keyword as a query, then it gives retrieved images such as apple computer or apple fruit related to user query. To improve image search result

    ,theonline image re-ranking is the efficient way. Re-ranking is an effective technique which is used by most of the image search engines.This is shown in figure 1input query is given by user,according to query keyword related images are retrieved by search engine.By selecting particular image from the images which are retrived by search engine,this reflects user search intension,images are reranked by visual similaries with the query image.Computation of images is carried out offline first and stored by search engine, And main online computation for reranking of images is carried out by comparing visual features.

    Sr. No.

    Author name




    Xiaogang Wang, KeLui


    Web Image Re-Ranking Using Query-Specific Semantic Signatures


    S.Vasukipriya,T. Vijaya Kumar 2013


    An Effective Recommendation By Diffusion Algorithm For Web Graph Mining


    S.Aarthi, Mr.S.Sampath


    A Heat Diffusion for Mining Web Graph for Recommendations using Recommendation



    Mrs.ShwetaShin dhe, Prof.



    Image Web Graph for Query Recommendation


    HauMa, lrwin King


    Mining Web Graphs for Recommendation


    Information Retrieval (CBVIR) which addresses the image retrieval problem that is the problem of searching for digital images in large databases. In CBIR the search engine analyzes the contents of the image rather than the metadata such as keywords, tags or descriptions associated with the image.A combining all types of visual features query technique that involves providing the CBIR system with an example image that it will then base its search upon. The underlying searchalgorithms vary depending on the application, but result images will all share common elements with the provided example.


    Query based search techniques have shown their in- effectiveness in obtaining image based results as there is mismatching of images and associated textual information resulting in irrelevant images appearing in the search results. The volatile growth of web information has created a crucial challenge for search engines to handle huge measure of data and has also increased the difficulty for a user to achieve his information. For example an image which is irrelevant to apple fruit will be mistaken as a relevant image if there is a word apple existing in the query which can otherwise mean apple computer. Various recommender systems are in use to overcome this existing problem of information overload. But the use of recommender systems makes the task of sorting images according to user intention more complex and needs user intervention to be fulfilled correctly. Various methods such as Context-Aware Recommender Systems (CARS) are in use to overcome this problem. Search engine giant GOOGLE uses a Content Based Image Retrieval (CBIR) method also known as Query by Image Content (QBIC) and Content Based Visual.

    In existing system, there are millions of web pages and not all of them are related to the users search intention and thus accurate results are not obtained for different users on the same device as more than one user can use the search engine. Threfore it is necessary to have profile for individual user so thatthe search results are improved according toindividual userintentions.However kind of systems are not developed for smartphone.


    1. Re-ranking precision

      For our new approaches, two different ways of computing Semantic Signatures are used. Query Specific Visual Semantic Space using Single Signature (QSVSS_Single)for an image, a single semantic signature is calculated from one SVM classifier which is trained byQuery Specific Visual Semantic Space using multiple signatures (QSVSS_Multiple).For an image, multiple semantic signature are computed for multiple SVMclassifiers each of which is trained on types of visual feature separately. Its working is shown in following example.

      Fig.2. Diagram of our new image re-ranking framework

      Example-The diagram of our approach is shown in Figure 2.The reference classes (which represent different semantic concepts) of query keywords are automatically discovered at the offline stage.If suppose we use a query keyword (e.g. apple), a set of most related keyword expansions (such as apple computer, apple fruit) are automatically got selected by considering both textual and visual information. This set of keyword expansions are having the reference classes for the query keyword which are already defined in them. The training example of a reference class, the keyword expansion can be obtainedautomatically.If suppose we used query keyword (e.g. apple) then images will be retrieved by the search engine. Imageswill be retrieved by the keyword expansions (apple) are much less diverse than those retrieved by the original keyword (apple).

      After automatically removingoutliers, the retrieved top images are used as the training examples of the reference class. Some reference classes (such as apple computer and apple fruit) have similar semantic meanings and their training sets are visually similar. The efficiency of online image re-ranking can be improved by removing redundant reference classes.For each query keyword, a multi-class classifier on low level visual features is trained from the training sets of its reference feature classes and stored offline. If suppose there are havingK types of visual features and if one could combine them to train a single classifier. It is also possible to train a separate classifier for each and every type of features.This shows that the re-ranking accuracy can be increased the latter choices but will also increase storage and reduce the online matching efficiency because of the increased size of semantic signatures.

      An image may be related to multiple query keywords. Therefore it might have several semantic signatures obtained in different semantic spaces. As per the word image index file, each image in the database is associated with a only few related keywords. foreach and every related keyword,there are lot many semantic signatures of the image are extracted by computing the visual similarities between theimage and the reference classes of the keyword using the classifiers trained in the previous step.

      The reference classes form the basis of the semantic space of the keyword. If an image has N related keywords, then it has

      N semantic signatures to be computed and stored offline.

      At online stage, a pool of images are retrieved by the search engine according to the query keyword input by a user. Since all the images in pool are related to the query keyword,they have pre- computed semantic signatures in the semantic space of the query keyword. If user chooses a query image, then re-ranking of all images is done by comparing similaries of semantic signatures.

    2. Profile based search engine

    Search is the most important feature available on the web for users. Popular search engines are available such as Google, Yahoo, Bing which provides output according to user intentions. Google uses a page ranking algorithm for providing output to users which is based on the keyword input by the users. However there are millions of web pages and not all of them are related to the users search intention and thus accurate results are not obtained for different users on the same device as more than one user can use the search engine. Thus a profile based system would solve this problem as every user would have his own profile and the semantic signatures of every users search term would be stored in individual profiles. Every user has to log in to their profile with a username and password to access the search engine. This way the search results are improved according to user intention.


In this paper proper indexing and image re-ranked is performed.on basis of user requiremenst or needs. The images are re- ranked using keyword expansion to provide better efficiency and effectiveness by using semantic signature for more precise output.Sowe propose profile based smatrphone image recommender which can be used in windows as well as in android platform for smartphone.Moreoverthere are millions of web pages and not all of them are related to the individual users search intention and hence accurate results are not obtained for different users on the same device as more than one user can use the search engine.Threfore profile based system can improve the search results according to individual user intentions.By using profile based system re-ranking of images is performed for fast image search for a particular user.


  1. Xiaogang Wang, Ke Liu and Xiaoou Tang, "Web Image Re- Ranking UsingQuery-Specific Semantic Signatures" IEEE Transactions on Pattern Analysis and Machine Intelligence (Volume: 36, Issue: 4) April 2014, DOI:10.1109/TPAMI.201.

  2. S. Aarthi and Mr. S. Sampath, A Heat Diffusion Method for Mining Web Graphs for Recommendations Using Recommendation Algorithm ISSN: 2278-0181, (IJERT) Vol. 2, Issue 8, August 2013.

  3. Mrs. Shweta Shinde1 and Prof.Dr.V.H.Patil, "Mining Web Graph For Query Recommendation" International Journal of Emerging Trends & Technology in Computer Science (IJETTCS),ISSN 2278-6856,Volume 2, Issue 3, May June 2013.

  4. Hao Ma, Irwin King, Senior Member, IEEE, and Michael Rung- TsongLyu, Fellow, IEEE,"Mining Web Graphs for Recommendations" IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING, VOL. 24, NO. 6, JUNE 2012.

  5. Fabrício D. A. Lemos, Rafael A. F Carmo, Windson Viana1 and Rossana M. C. Andrade1, "Towards a Context-Aware Photo Recommender System".

  6. P. Hong, Q. Tian, and T.S. Huang, Incorporate Support VectorMachines to Content-Based Image Retrieval with Relevant Feedback,Proc. IEEE Intl Conf. Image Processing, pp. 750-753, 2000.

  7. W. Niblack, R. Barber, W. Equitz, M. Flickner, E. Glasman, D.Petkovic, P. Yanker, C. Faloutsos, and G. Taubino, The QBICProject: Querying Images by Content Using Color, Texture, andShape, Proc. SPIE Storage and Retrieval for Images and VideoDatabases, pp. 173-181, 1993.

  8. M. Sudhamani and C. Venugopal, Image retrieval from databases: An approach using region color and indexing technique, Int. J. Computer Science and Network Security, vol. 8, no. 1, pp. 54-64, Jan. 2008.

  9. J.C. Burges, A Tutorial on Support Vector Machines for Pattern Recognition, Data Mining and Knowledge Discovery, vol. 2, no. 2, pp. 121-167, 1998.

  10. M. K. Mandal, T Aboulnasr, and S. Panchanathan, Imageindexing using moments and wavelets," IEEE Transactions on Consumer Electronics, vol. 42, no. 3, pp. 557- 565, Aug 1996.

  11. J. Cui, F. Wen, and X. Tang. Intentsearch: Interactive on-line image search re-ranking. In MULTIMEDIA 08: Proceedings of the 16th annual ACM international conference on Multimedia, 2008

  12. L. Zhang, F. Lin, and B. Zhang, Support Vector Machine Learning for Image Retrieval, Proc. IEEE Intl Conf. Image Processing, pp. 721-724, 2001.

  13. N. Ben-Haim, B. Babenko, and S. Belongie, Improving web- based image search via content based clustering, in Proc. Intl Workshop on Semantic Learning Applications in Multimedia, 2006.

  14. Natsev, R. Rastogi, and K. Shim, WALRUS: A similarity retrieval algorithm for image databases, IEEE Trans. Knowl. and Data Eng., vol. 16, no. 3, pp. 301-316, Mar. 2004.

Leave a Reply