A segment-based advertisement search method from TV stream

Searching of video clip from TV streams has significant commercial values, for example, the application of advertisement tracking in different TV channels. In this paper, a segment-based advertisement searching method is proposed. Robust visual features are first extracted to overcome some video transformations, and then two search strategies are presented for long AD clips and short ones. The experimental results indicate that the average search time of one query clip from a 24-hour TV stream is about 1 second, and the mean recall for 9 channels exceeds 98% while 100% precision is achieved. The evaluation results of copy detection task at TRECVID show the robustness and effectiveness.