Font Size: a A A

Some Thesis Research On Content-based Image Retrieval

Posted on:2004-08-22Degree:DoctorType:Dissertation
Country:ChinaCandidate:A X HongFull Text:PDF
GTID:1118360095961713Subject:Applied Mathematics
Abstract/Summary:PDF Full Text Request
With rapid development of internet and technologies of computer and multimedia, information of all kinds of images becomes more and more great. It is very important to organize, manage and utilize these multimedia information. Until now, there are still many problems of this research area to be solved. This thesis focuses on the technology of content-based image retrieval (CBIR) and get some achievements in this area.Based on rich introduce of image features, similarity matching and indexical technologies, methods of evaluation were investigated and several representative image retrieval systems were introduced. Afterwards, the current difficulty of CBIR, the range and significance of the research were pointed out.One of short points of traditional color histogram is lack of spatial information .To overcome this shortcoming, the space color histogram was introduced in this thesis and a correlation measure between color bins was proposed which considered the correlation between colors. A hybrid measure based on space color histogram and correlation measure was introduced. Experimental results prove that hybrid measure is better than Lp measure.In this article, based on the introduction of framework of CBIR with relevance feedback, a real image retrieval system was constructed. As we know, retrieving image through semantic features is the most desirable and useful way. In this thesis, a semantic feature databases built dynamically based on relevance feedback was proposed. During the retrieval process, the user's perception subjectivity is captured so as to realize retrieving imagf by semantic feature.By analyzing the correlations of r, g, b components of a color image a new color image compression coding method based on fractal theory was proposed. Compression ratio was increased and coding speed was fairly high.Based on adaptive quad-tree fractal image encoding scheme, the concept and algorithm of peak signal noise ratio (PSNR) based on the fractal codes is put forward, and is used to match the similarities between images. Experiments show that the matching scheme in this paper is much more close to human perception, and this technology is very suitable to construct and manage huge image databases.In this thesis, we propose an efficient segmentation method based on color clustering and domain knowledge-guided to extract flower regions from flower images. Experimentsshow that our flower region extraction approach based on color clustering and domain knowledge-guided can capture accurate flower regions. The retrieval results show that our Region-Of-Interest (ROI) based retrieval approach performs much better than the Swain's method based on global color histogram.
Keywords/Search Tags:Content-based image retrieval, Hybrid measure based on color histogram, Relevance feedback, Semantic features database, Peak signal noise ratio based on the fractal codes, Image databases.
PDF Full Text Request
Related items