A Video Annotation Method Based on Color Statistics

The development of automatic video annotation technique is still in the primary stage compared to the image automatic technique. The existing methods have many drawbacks in the universal applicability and efficiency. Against to these existing problems, this paper proposes a new video annotation technique based on the RGB and Lab color space. This method structures a 4096-dimensional feature vector which can characterize video content, respectively in two color spaces, and computes the distances between the feature vector of the target video and that of the labeled videos from the video library. Then it transfers the proper label to the target video using a simple greedy algorithm. The experiment results demonstrate that our method is effective for videos with simple situation, and also makes great progress for videos with complex situation.