Study Of The Content-Based Video Classification Algorithm

With the rapid development of computer networks and the multimedia technology, there occurs more and more multimedia information. As a common multimedia, video is closely bound up with our life.Because the number and the kinds of the videos increase steadily, packing up the enormous videos and obtaining the content of interest have become an important research issue. Therefore, the automatic video analysis system emerges as the times require, and the research of automatic video classification algorithm also becomes a research focus.Based on the analysis of video classification algorithm, this paper sums up the available algorithm and analyzes the merit and the drawbacks in them. Then the cartoon video classification is chosen to initiate the research. By analyzing the difference of visual content between cartoon and other videos, a new scheme is proposed for describing the video features. Many features from MPEG-7 visual descriptors and image quality assessment measurement are used in this scheme. Finally, 8 typical features are extracted to distinguish the cartoon videos.At the same time, a new content-based video classifier is designed by incorporating the support vector machine (SVM) algorithm and the active relevance feedback technique. Compared with the classifier based on relevant feedback technique and SVM, the proposed new classifier not only can obtain higher classification accuracy, but also cost fewer iterations.The experimental results on the vast real video clips illustrate that the proposed scheme of video feature expression and the new classifier design method obtain a higher accuracy. Specially the proposed classifier has a higher advantage of cartoon video classification, compared with the classifier based on SVM and that based on traditional relevant feedback technique and SVM.