摘要：Perceptual coding should take full advantage of the results from human visual system studies. Image can be parsed into two main categories of representation: contour curve and region texture. In this paper, we introduce a multi-regions texture substitution (MRTS) algorithm for images and videos. MRTS algorithm can achieve medium-level-visual texture substitution through selecting the smallest region texture sample. Specifically combining with dynamic texture learning and synthesis techniques, we explore it in 2D and 3D to perform multi-regions texture substitution for video sequences. Experimental results confirm MRTS algorithm can be done successfully for images with massive textures. Compared with JPEG2000 compression, MRTS can not only get more coding gain but remain most image information people care. MRTS algorithm also can get good video reconstruction quality.
