Research On Multi-modal Pornographic Video Detection Algorithm Based On Unsupervised Features And System Implementation

With the development of the Internet,more and more pornographic images and video resources flood on the net,which brings adverse effects to people,especially to teenagers.An automated and efficient method of pornography detection is necessary.In this paper,unsupervised learning is introduced to pornographic video detection.An algorithm of multi-modal pornographic video detection based on unsupervised features are proposed.The principal work and remarks of this paper are as follow:(1)For static video frames,a detection algorithm based on convolutional linear decoders is proposed.Compared with traditional skin-based methods,the image features learned can find some structural information which is useful for detection.(2)For video sequences,a detection algorithm based on statcked independent subspace analysis is proposed.Combined with bag of words model,the motion features outperform traditional hand-designed motion features since they are not sensitive to video format and catch more useful local patterns.(3)The framework of multi-modal detection is established,which combines image features,motion features and audio features.(4)A new porn video library is built,which contains different types of pornographic videos to compare different algorithms.