A new approach to disparity compensation for stereo image compression is proposed based on generalised block matching. The proposed method results in a better picture quality which is more pleasing to the human viewer due to the elimination of the blockiness effect and the compensation of occluded areas. It is shown that the disparity estimation between the left and right eye picture can be improved by at least 3 dB compared to the conventional full search block matching, by using nonlinear mapping functions to describe the co-ordinate correspondence of related pixels in the two pictures.
Read full abstract