Object-oriented coding using successive motion field segmentation and estimation

Block-based motion compensation assumes that all pixels within a block have the same translational motion. That hypothesis, however, results in inaccurate compensation of moving objects´ boundaries. Object-oriented video compression algorithms typically segment each image in regions of uniform motion and estimates the motion of these regions to generate more accurate motion compensated images. We present a two-stage algorithm for motion field segmentation and estimation in an object-oriented coder. In the algorithm´s first stage, a standard block-matching algorithm and a maximum a posteriori probability estimate are used to compute a translational motion field and its segmentation. This segmentation is then utilized in the second stage to estimate the parameters of complex motion models. The parameters of the complex motion models are only estimated in the algorithm´s second stage which reduces the computational complexity of the proposed algorithm. Simulation results show that the proposed algorithm significantly reduces the bit rate needed to encode video sequences when compared to standard block-based algorithms