Efficient Video Object Classifier using Locality-Enhanced Support Vector Machines

IEEE Press
Publication Type:
Conference Proceeding
IEEE- International Conference in Systems, Man and Cybernetics (SMC), 2004, pp. 6373 - 6377
Issue Date:
Full metadata record
Files in This Item:
Filename Description Size
Thumbnail2006007964OK.pdf365.6 kB
Adobe PDF
In multimedia applications such as MPEG-4, an ejicient model is required to encode and classih @ideo objects such as human, car and building. Recently Su,?port Vector Machine (SVM) has been shown to be a good c!assijer: however; its larxe computational requirementpmhibited its use in real time video pmcessing applications. In this paper: a modelispmposedthatenablesuseofSVMin video applications. Thispaper aims IO merge multi-scale basedxelecfive encoding/classification technique and locality-enhimced Support Vector Machine (SVM). Thepmposed model allows selected image scales (of interesr) ro be encoded and classified more accurafely by complex classifier such as ,SVM, whilst other image scales of less significance IO be encoded and classified by simpler encoder/classifiex Image sca Ies of interest are readily selectedfioni niulti-scale image processirigparadigm. SVMis used to encode visual object infirmation of significant image scale only: hence its use is ejicient. Experiment with MPEG-4 video object encoding and classification shows that the performance of the proposed ?!lode1 is comparable with other models, however with significarifly reduced computational requirements.
Please use this identifier to cite or link to this item: