DC Field | Value | Language |
---|---|---|
dc.contributor.advisor | Lee, Heung-Kyu | - |
dc.contributor.advisor | 이흥규 | - |
dc.contributor.author | Park, Gun-Han | - |
dc.contributor.author | 박근한 | - |
dc.date.accessioned | 2011-12-13T05:20:43Z | - |
dc.date.available | 2011-12-13T05:20:43Z | - |
dc.date.issued | 2004 | - |
dc.identifier.uri | http://library.kaist.ac.kr/search/detail/view.do?bibCtrlNo=237662&flag=dissertation | - |
dc.identifier.uri | http://hdl.handle.net/10203/32859 | - |
dc.description | 학위논문(박사) - 한국과학기술원 : 전산학전공, 2004.2, [ x, 106, v p. ] | - |
dc.description.abstract | Rapid advances in science and technology have produced a large amount of image data in various areas. We often need to store, manage, and retrieve image data to perform assigned tasks and to make intelligent decisions. Content-based image retrieval (CBIR) methods have been developed in recent years for distinguishing useful images from extraneous ones, and for retrieving relevant images that satisfy user interests. In general, a CBIR system extracts visual features from an image, transforms the image into a feature vector, and calculates the similarities between a query image and images stored in the database. Then, it presents a sequence of images in decreasing order of similarity by ranking. Users evaluate the effectiveness of a CBIR system by the ranked results, but the relevant images are often not at the top of the rankings. Even if a system finds particular relevant images, many others may be substantially further down in the ranking. To remedy this problem, more accurate retrieval methods based on analyzing the visual features are required. In this thesis, we propose a retrieval method using post-retrieval clustering for CBIR. We classify the retrieved results into sub-groups via a post-retrieval clustering method. The sub-groups so formed should be such that members of the same sub-group have a high degree of association to each other and a low degree of association to members of other sub-groups. Thus, when images are re-ranked according to the proposed algorithm, the images of the cluster that is nearest to the query image could achieve a rank higher than they did in the originally calculated rank. In addition, we analyze the effects of clustering methods, query-cluster similarity functions, and weighting factors in the proposed method. We conducted a number of experiments using several clustering methods and cluster parameters. Experimental results show that the proposed method achieves an improvement of retrieval effectiveness of over 10 % on averag... | eng |
dc.language | eng | - |
dc.publisher | 한국과학기술원 | - |
dc.subject | RE-RANKING ALGORITHM | - |
dc.subject | VISUAL FEATURE VECTORS | - |
dc.subject | CLUSTERING | - |
dc.subject | IMAGE RETRIEVAL | - |
dc.subject | SIMILARITY RELATIONSHIP | - |
dc.subject | 유사도 관계성 | - |
dc.subject | 재순위 알고리듬 | - |
dc.subject | 시각특징 벡터 | - |
dc.subject | 클러스터링 | - |
dc.subject | 이미지 검색 | - |
dc.title | Effective image retrieval based on clustering of visual feature vectors | - |
dc.title.alternative | 시각특징 벡터의 클러스터링을 이용한 효과적인 이미지 검색 방법 | - |
dc.type | Thesis(Ph.D) | - |
dc.identifier.CNRN | 237662/325007 | - |
dc.description.department | 한국과학기술원 : 전산학전공, | - |
dc.identifier.uid | 000965144 | - |
dc.contributor.localauthor | Lee, Heung-Kyu | - |
dc.contributor.localauthor | 이흥규 | - |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.