Bitgroup modeling of signal data for image compression

Summary form only given. Binary variable order adaptive algorithms like the UMC of Rissanen (1986) and JBIG can be used to losslessly compress non-binary data by splitting the data into planes, each of 1 bit resolution, and passing each plane to a separate instance of the algorithm. The UMC algorithm operated in this way is the most powerful lossless signal data compressor the authors are aware of. We attempt to develop an understanding of why this approach is so effective. We investigate the common technique of Gray coding the data before splitting it into single-bit planes and passing to the modeler and coder, and compare it to a simple weighted binary coding. We then propose a non-binary pseudo-Gray code as a method of generating planes of resolution greater than or equal to 1 bit, and compare it with the other conventional methods. The algorithm to generate the pseudo-Gray code is much the same as that for the construction of a binary Gray code, except that instead of minimizing the Hamming distance between neighboring bit planes, we instead minimize the Euclidean distance between adjacent groups of bit planes.