Iterative Gradient Method for Movement Estimation
Masahiro Wada, Hirohisa Yamaguchi · Systems and Computers in Japan · 1986
Abstract In the processing and coding of moving images, the movement of the image is one of the most important parameters. In the realtime processing, it is desirable that the movement be estimated with less computation and simple hardware. From such a viewpoint, this paper proposes a movement estimation method with less computation and satisfactory estimation accuracy. In the proposed method, the gradient method is applied iteratively to the pixel blocks, improving the accuracy of estimation while retaining the merit of the gradient method. This paper describes first the principle of the gradient method, indicating the deterioration of the accuracy. The trend is demonstrated by experiment. Then the principle of the proposed method is described, and it is shown by theoretical discussion and simulation experiment that the accuracy of estimation can be improved. Finally, from the standpoint of motion‐compensation coding, a comparison is made with other methods based on the interframe error entropy. As a result, it is shown that estimated movement among adjacent blocks can be utilized in the proposed method, and a motion‐compensation performance can be obtained which is as stable as in the matching method, which requires a larger amount of computation.