Exploiting local colour dependencies in binary tree predictive coding

Binary tree predictive coding (BTPC) is a general-purpose compression scheme for still images. On its initial release, two problems were identified with BTPC: the most visible coding artifact at low data rates is contouring. This effect should be reduced, even at the expense of objective quality (PSNR). BTPC performs significantly poorer than GIF for limited-palette, highly-structured colour graphics, where colour quantization allows very high compression. Today, the most telling problem for BTPC is that the research systems against which it was compared have been superseded by standards. This paper reports developments aimed at solving the above technical problems and updated experimental comparisons showing that the latest version of BTPC remains competitive for general-purpose image coding.