A Novel Method to Classify Videos Based VBR Trace

摘要

Video classification research has been studied for many years. Traditional video classification methods are based on text, sound, and visual content. However, all these approaches require that the video content can be inspected. If the video content can not be investigated. For example, the video frame is encrypted or transmitted on the network device, then we can only measure the size of the video frame bitrate. In this paper, we propose two novel feature extraction methods based on variable bit rate (VBR) trace. The first one is extracting features in sliding windows. The second one is based on change points techniques to obtain more reasonable windows. We carry out empirical studied on our data sets to discriminate the action videos from the other videos. The experiment shows that we can identify the action video with 87% g-mean.