Parameterized Markov models for efficient compression of grayscale images

Lossless compression using finite context variable order Markov models generally achieves smaller compression ratios for small and medium sized images than for text data of the same size. This is due to (1) the larger alphabet size, (2) the enormous number of different contexts especially in higher order models, and (3) quantization noise introduced during the digitization process. Theoretically, Markov models will eventually capture the characteristics of the image data provided there is enough data. In practice, there are hardly any images of appropriate size. Therefore, to improve the compression ratios for images, four image-specific and two model-specific techniques to parameterize finite context variable order Markov models are proposed