Font Size: a A A

Research On Content-based Image Retrieval For Universal Techniques And Applications

Posted on:2005-11-05Degree:DoctorType:Dissertation
Country:ChinaCandidate:D Y XiaFull Text:PDF
GTID:1118360152469049Subject:Communication and Information System
Abstract/Summary:PDF Full Text Request
Aimed at the increasing application demands of multimedia information retrieval onInternet and the defects existed in content-based image retrieval (CBIR) systems forgeneralization design and performance optimization, in the dissertation, some importantquestions of CBIR for universal techniques are deep discussed and a set of integratedsolution is presented, in which some new methods of image feature description, extraction,matching, index, comparability measure and fast retrieval are researched for constructing auniversal model of CBIR. A great deal of test is done in image database and on Internet, andthe rationality of the solution is validated by experimental results. At first, the background and signification, development and application of CBIR foruniversal techniques are summarized. Then, four aspects of main questions are researched,which includes image feature analysis, image retrieval technique, image search technique,system design and application. In section 1, some typical of image features in CBIR systems,such as color, texture, shape, etc., are detailedly analyzed, and a new method combinedwavelet with relative moments is presented, in which image characteristics of verge andregion are integrated, the feature descriptions of verge, region and structure are unified, andthe invariability of move, zoom and spin is met for improving the existent shape descriptionmethods and still more according with human vision. In section 2, two types of imageretrieval techniques in image database are researched, some key techniques, methods andimprovement measures of CBIR are discussed, and a universal image retrieval model basedon Bayesian classification and interactive relevance feedback between human and computeris presented, in which low layer features based on machine vision and high layer featuresbased on semantic description are combined, different retrieval algorithms and new retrievalfunctions are easily replanted and extended for meeting the demands of image classificationand retrieval in large image database or network database. In section 3, tow types of imagesearch techniques on Internet are researched, some key questions of content-based imagesearch (CBIS) techniques in the processes of practicality are explored and two kinds ofsolution are presented for meeting the application demands of different users. Solution 1 isthe combinations of text-based image search techniques on Internet and CBIR techniques inlocal image database, and the processes of text-based image search and CBIR areasynchronously implemented. The solution is simple and convenient and may partly meetthe application demands of users. Solution 2 is the perfect combinations of CBIS and CBIR IIItechniques. In the solution, the processes of CRIS and CBIR are synchronouslyimplemented, the extraction and matching of image features are completed on Internet, andboth search time and retrieval outcomes may be accepted by common users. In section 4,some main questions of CBIR for universal techniques in system design and application areexplored. Based on the research fruits above, a universal structure of CBIR system isdesigned and some optimization measures of system performances are presented for solvingsome key problems in the processes of practicality, which includes the optimizationmeasure of system interface based on MPEG-7, the optimization measure of database basedon the classification method of clustering and Bayes, the optimization measure of hardwaresystem embedded with DSP and the optimization measure of distributing search strategybased on multi-Web Crawler in parallel work. Specially, the optimization measure ofhardware system embedded with DSP is presented for improving system efficiency. Besides,aimed at the existent performance estimate method of CBIS engine on Internet andcombined the performance test results of our system, an improvement performanceestimation method is put forward for enh...
Keywords/Search Tags:content-based image retrieval, vision feature, feature extraction and matching, Web-based information collection, image classification, relevance feedback, universal model, performance optimization
PDF Full Text Request
Related items