Obscene video detection using mutiple-classifier fusion

Obscene video detection is a core technology to prevent inappropriate access of children or teenagers to obscene video contents. There are many obscene video or image detection methods such as skin region analysis based methods, global histogram based methods. However, accuracy of these methods are not high enough to be deployed in the real-world environment. In this paper, we propose an obscene video detection method by multiple-classifier fusion. Three fusion methods are proposed: precision-oriented, recall-oriented, and accuracy-oriented. Experimental results show that by using the multiple classifier fusion method, superior accuracy, precision and recall can be achieved while exploiting the complementary behavior of different obscene classifiers.