A region-based blurring algorithm to reduce bitrate in very low bitrate video coding is proposed. This algorithm reduces the bitrate by passing the original background image through a filter before motion estimation. Each frame in original sequence will be segmented into foreground and background at first. The background part, which is of less importance for human vision, will be blurred through a blurring filter, while the foreground is kept unchanged. The bitrate for head-and-shoulder type sequences can achieve 5%/spl sim/20% bitrate saving by this blurring algorithm. According to the human vision model, PSNR degradation in the background is not evident. Using this blurring algorithm, but keeping the coded bitstream at the same bitrate as the original sequence, about 0.5/spl sim/1 dB gain in the foreground can be achieved due to bits reallocation between foreground and background.
展开▼