To improve the encoding efficiency and reduce arithmetic processings by selecting one kind for each block out of two kinds of block image and their means block image.
Local movement compensation is performed between a source image 401 and a reference image 410 to find a local movement compensated block image 412 for a block 402 in the original image 401. At the same time, global movement compensation is performed to find an image in a block at spatially the same position with a block 402 in a composited global movement compensated predictive image 420 as a global movement compensated block image 412. The mean block image 440 is generated by averaging pixels of the local movement compensated block image 412 and global movement compensated block image 421. The image having the best encoding efficient is selected out of those three block images 412, 421, and 440. Block images selected by blocks in the source image 401 are gathered to composite a predictive image 460 as an object of predictive error encoding with the original image 401.
NAKAYA YUICHIRO