画像のアフィン変換動き補償予測に関する検討

Abstract

The motion compensated prediction is an essential technique for redundancy reduction coding of video sequences. This paper describes motion compensated prediction based on affine motion model where not only translation motion but also rotation and scaling motions are compensated. Computer simulation experiments using actual video sequences were carried out in order to clarify the relationship between affine parameters and coding performance. Prediction error entropy and affine parameter entropy were estimated for various parameter sets. The result have shown that half-pel translation is most effective for reduction of total coding entropy and that rotation and scaling are less effective compared with the translation.