Semantic video retrieval system based on SVM
Yibo Jiang · Jisuanji gongcheng yu sheji · 2010
To narrow the semantic gap,the application of support vector machine(SVM) in video semantic retrieval is studied.Firstly the support vector regression(SVR) is used to realize semantic auto annotation.Then a semantic retrieval with user feedback based on SVM classifier is constructed,after which the conventional relevance feedback approaches is improved.The train samples is increased dynamically and remain in balance with accumulating the samples,and optimizing the selection of negative samples as well as balancing the positive samples and negatives ones.This feedback information will be shared in the next same search with storing the SVM model when a satisfied retrieval.This we called the long-term memory mechanism of SVM relevance feedback.Experimental results show that,this improved semantic retrieval of video frame is significantly more effective than the content based retrieval that also using SVM feedback mechanism.