A multimode context-based lossless wavelet image coder

Summary form only given. Currently, the most difficult part of a context-based lossless image compression system is how to determine the contexts. There is a tendency that the techniques for choosing contexts are based on practical experiences. For example, Wu et al. proposed the gradient of the current pixel as the base to choose the contexts for CALIC. Weinberger et al. proposed a similar strategy for LOGO-I. The main idea of this paper is to establish a general concept for the context selection mechanism. One way to do it is to employ an adaptive system that can switch between context schemes if there is enough evidence that switching will give us some performance gains. To be able to do this, some statistics such as the message length for each context scheme and the occurrence of each scheme-have to be collected along the way. Given several possible context selection schemes, the proposed system chooses the scheme that provides the best total message length in the immediate past. The first objective of this switching mechanism is to improve the compression even further. This new concept is combined with wavelet transformations to form a new lossless image compression system. Golomb-Rice code is used to encode the symbols, to reduce the overhead in updating the statistics and actually encoding the symbols. The current results of the experiments indicate that switching can improve the performance only slightly. This is due to the fact that each possible context performs almost the same. The overall results are slightly worse than CALIC or LOGO.