Image compression using projection vector quantization with quadtree decomposition

This paper addresses a new image compression algorithm that employs the quadtree decomposition and the projection vector quantization (PVQ). We examine the relationship between the image object and its projection data and discuss the image reconstruction problem from limited data. By adopting the quadtree decomposition, we can segment the image into variable sized blocks which can be well represented by the single orientation reconstruction (SOR). While the proposed algorithm can easily deal with larger blocks than the conventional VQ techniques, the bits can be allocated adaptively according to the information contents of the image. Also, the proposed algorithm greatly reduces the complexity associated with the recursive transmission of the PVQ. In addition, it shows improved performance both in subjective and objective quality compared with the case that uses fixed sized blocks.