Content-Based Classification For Web Video

With the rapid growth of data transmitted in the Internet, and the progress of compression storage technology, the number of Internet videos have has exploded over the last few years. In order to manage video data effectively, video content-based classification techniques emerged. Content-based video classification technology plays an important role in video annotation, retrieval and filtering on specific information.This paper focuses on the classification of video genre. The performance of video classification mainly depends on the selection of feature and classification method. From the perspective of video content this paper studied the differences between the different types of video content and proposed audio-visual based feature extraction algorithm. In the extraction of the audio information, we use a block-level representation of audio features, better expressed audio time domain and frequency domain information. In the extraction of video information, we first use a variety of features to achieve the video shot segmentation; then focused on the human visual perception to extract the relevant information from the video shot and color. In shot-based feature extraction, we use rhythm, action and some other feature; The color feature extraction mainly include statistical distribution of color, basic color, color characteristics, color relationship these four aspects. In the multi-classifier selection, this paper use a directed acyclic graph of SVM multi-classification. Finally, this paper collected seven categories of video from the Internet, four groups of classification experiments conducted to verify the effectiveness of the classification algorithm and achieved a good classification results.