An Audio-Based Method For TV Program Extraction And Analysis

As the amount of digital video content (TV program data in particular) is increasing rapidly on the Internet, the demand for many new online services such as TV-on-Demand, grows fast at present. So we need an efficient method to extract the TV program from the content-unknown massive TV data and do further content-based video parsing work. In this paper, we summarize our research work on audio-based TV program extraction and analysis in the framework of "content-based video retrieval"On the TV program extraction, we propose a duplicate detection method for TV macro segmentation, which attempt to find all inter-program to achieve chaptering. Then the reference audio template, which is learned in duplicate detection phase, is used to detect the specific TV program. In our research, the innovation is that audio fingerprinting technique is brought in to perform duplicate clip matching, and computer vision technique is adopted to extract the feature of audio clip. The experiments demonstrate that the approach presented can reach the state of the art.On the TV program analysis, an audio segmentation and classification approach for news program is presented. The presented system segments the news audio into clips and then classifies the clips into four classes, which are silence, speech, music and noise. The experiment results are satisfying, and prove the effectiveness of the proposed method.In this paper, some technique and theoretical results of the summarized research work are also valuable for the other research work in the area of "content-based video retrieval".