An image coding method by prediction of multiresolution subimages

This paper presents a novel coding technique using the edge information and the correlation among multiresolution and multi-frequency subimages of an original image. To obtain a high compression ratio without visible subjective degradation, our scheme separates the coding of the strong transitions (or edges) and the residue source component. The edge source is coded by a chain coder. The residue source is coded using multiresolution subimages generated by applying a subband filter. Each block in the subimages is predicted by an affine transformation of blocks in lower resolution subimages with the same orientation. Since the proposed scheme is a non-iterative scheme, the affine transformation does not compose any contractivity constraint on the block mapping or isometry operation. Experimental results show that the performance of our scheme is superior to the iterated transformation theory-based coder and other popular image coders such as JPEG in terms of both the visual quality and bit-rates.