Skip to Main Content
In this paper, we present our method for effective and efficient classification of different types of video that make use of the low level visual features, color and motion features. The color features, similar to the MPEG-7 scalable color descriptors, are the compressed color histograms extracted from the DC images of the video sequences. The motion features, similar to MPEG-7 motion activity descriptors, are extracted from the motion vectors. Both features are extracted from the compressed domain and give a good characterization of the video in both the spatial and temporal dimension. We then classify different types of video using hidden Markov model. First, we train the hidden Markov model for each type of video. Then the trained HMMs are used to classify incoming videos using maximum likelihood classification.