Region-Based Motion Compensation for Video Compression using a Quadtree Approach

In this paper, a new motion field segmentation algorithm using a quadtree structure is presented for use in video compression systems. The goal of motion segmentation is to obtain a motion field representation that more fully exploits 2D redundancy in motion fields than is possible with block-based schemes, thus requiring fewer bits to code. It is shown that the new method achieves the same or similar MSE as conventional block-based motion estimation, while significantly reducing the number of motion bits. The number of bits needed to code the resulting DFD is not significantly increased.