Video Image Hierarchical Classification System in Compressed Domain Using SVM

カテゴリ: 研究会(論文単位)

論文No: IS10006

グループ名: 【C】電子・情報・システム部門 情報システム研究会

発行日: 2010/04/22

タイトル(英語): Video Image Hierarchical Classification System in Compressed Domain Using SVM

著者名: Cheng Yu(Tsinghua University),Zhang Tao(Tsinghua University),Guo Xuejing(Tsinghua University)

著者名(英語): Cheng Yu(Tsinghua University),Zhang Tao(Tsinghua University),Guo Xuejing(Tsinghua University)

要約(日本語): Video image classification becomes more and more ubiquitous and importamt as the volum of captured video is verylarge. Therefore, tools for classification and searching video databased are indispensable. Current clustering techniques thatextract features purely based on the whole image are struggling to achieve good results. This paper presents a novel face-basedimage clustering system with fuzzy classifier based on Support Vector Machine, mainly for compressed MPEG-1, MPEG-2 videoor JPEG image sequence. Our paper introduces two advantages: first, unlike traditional SVM classifier, usually treating binaryclassification problem, our method dynamically build a fuzzy hierarchical structure from the training data and cluster the similarimages into group of multiple class. The second is, the proposed system use features mainly from the human face to reduce thesearching place. In order to evaluate the performance of the proposed method, different image sequences were used. Performanceevaluations have demonstrated effectiveness of our method with high accuracy. The system can be extended to no-compressedvideo image sequence as well for various applications.

原稿種別: 日本語

PDFファイルサイズ: 583 Kバイト