Compensation methods using signal processing and adaptive quantization for better mean shift tracking on compressed video

In this paper, we discuss methods for coding MPEG-4 video to enable better computer vision on the compressed video. This problem is important since the amount of video content being generated from a variety of sources is increasing very rapidly, almost all this video is in compressed form, and computer vision applications such as video analysis, mining and querying running on these videos are becoming more commonplace. If the particular computer vision algorithm that is likely to be run on the compressed video is known apriori, then steps can be taken during the encoding process to facilitate the performance of the algorithm. In this paper, the algorithm we focus on is Mean Shift tracking, a robust and widely used tracker. We show that by performing signal processing on the input signal before it is encoded, or by adaptively changing the parameters of the encoding process, we can make the resulting signal more robust to degradations in the encoding process. The result is better tracking on the compressed video, at the same bitrate, but with some loss in PSNR.