Font Size: a A A

Research On Content-based Trademark Image Retrieval Technology

Posted on:2012-03-12Degree:MasterType:Thesis
Country:ChinaCandidate:J B SongFull Text:PDF
GTID:2218330368497578Subject:Signal and Information Processing
Abstract/Summary:PDF Full Text Request
With the development of information and network technology, a large variety of visual content is generated. Image as an important component of the visual content, it is able to image the expression of information directly .How to retrieve images from a large number of the contents of users' own interest, that is, an effective information retrieval, has become a hot issue .As a special image information carrier concerned, the mark is not only valuable in itself, but also has the ability to create value. It is one important industrial property to a company and is a valuable social asset. In recent years, with the increasing of the number of registered trademarks, it is so hard to compare images artificially that the content-based image retrieval CBIR came into being. In this paper, firstly depth study about the key technologies of the CBIR system, based on analysis of existing research, combined with BIC-pixel analysis algorithm proposed the level tree matching method LTM, then according to the LTM and BIC proposed the relevance feedback algorithm and using the method proposed design a CBIR system.This paper has the following contribution:(1) Further study on the underlying characteristics of image, combined with the trademark's own characteristics, focused on the color characteristics, and summed up the general extracting method of the underlying characteristics.(2) Research on the development status of the CBIR systems home and abroad, summed up the model of CBIR system, in order to compare the system performance, proposed the system evaluation criteria.(3) BIC image analysis algorithm for further study, compare the dLog distance function proposed in BIC algorithm with the vector distance analysis within the experiments.(4) Based on BIC-pixel analysis algorithm proposed the tree matching method level LTM, by this method the performance related to trademark image retrieval greatly improved, and it can reduce the image storage space effectively.(5) Combining the BIC algorithm with the method LTM, relevance feedback algorithm RF is proposed, and design the experiment system, compare the feedback algorithm RF with the traditional algorithm Rocchio , the result of experiment proved that the RF algorithm is better to improve the system performance of retrieval accuracy.Finally, analyze and summarize the study work the further research work is prospected.
Keywords/Search Tags:Content-based image retrieval, image feature, image pixel classification algorithm BIC, Level Tree Matching LTM, Relevance Feedback algorithm RF
PDF Full Text Request
Related items