SHANGHAI JIAOTONG UNIVERSITY(SJTU-IS) participation in high-level feature extraction at TRECVID 2009

In this paper, a description of our participation in high‐level feature extraction at TRECVID 2009 will be given. Four runs had been submitted for our effort on the task. For those four runs, they have some common points and are based on one system. The basic elements for our work is different kinds of the descriptive of any frame(picture) retrieved from the videos. We have taken six kinds of image descriptive into consideration this time, which are sift‐bow, local binary pattern, color moment, color hist, edge hist and gabor. Then we tried the different combination of those descriptive in the training and classifying of SVM. The result of the task is based on the linear weighted fusion of the separate result of each descriptive. We submitted the following four runs: