Objectionable Video Filtering Using Hierarchical SVM Classifier

Abstract

As the P2P software prevails on the internet, people contact with the objectionable information＊ more often than before. Because the objectionable information is not suitable for the minors, how to block or filter the objectionable information has became a critical issue. One of the major objectionable information is pornographic videos. Many studies have been researched on filtering objectionable images, but few studies have been investigated on filtering objectionable videos. In this paper, we propose a high accuracy objectionable video classifying system. We extract frames from videos to classify objectionable videos with a two-tier SVM classifier. In the first tier, we adopt the traditional image classifier to classify video frames. In the second tier, we propose methods to analyze the classification results from the image classifier in the first tier and generate features to classify videos with a second tier SVM classifier. We show that even if the image classifier in the first tier is far from perfect the proposed two-tier classifier can still produce satisfactory result in classifying videos. Finally, our experiment results suggest that the proposed methods are promising and applicable in real world situations.