Human subject-based video browsing and summarization

To acquire digital videos is much easier than before, since we can get videos captured from DV camcorder. More video archives make searching the targeted content more difficult. In the past decade, efficient video indexing, browsing and summarization techniques thus have become an important research issue in the field of content-based video retrieval. In this work, a novel mechanism of human subject-based browsing and summarization is proposed. Human subjects who are actually watching towards the camera are first detected and are further recognized by our proposed online learning classifiers, which are based on the measure of Mahalanobis distance. A complementary set of features is extracted from the torso of a human subject, as that part of the body contains relatively richer information than the face when the videos are captured in general camcorders. The features extracted from the torso consist of the quadtree-based color features and the quadtree-based edge features. Our experimental results show the efficacy of the proposed system for the human subject-based video browsing and summarization task.