Compressing color-indexed images with an adaptive palette reordering method

Abstract

Color-indexed images are widely found in various image applications nowadays. An efficient compression algorithm for coding color-indexed images can help to reduce their data size for saving both transmission bandwidth and data storage requirement. A color-indexed image is represented with a color index map each element of which serves as an index to select a color from a predefined set of colors called palette to represent the color of a pixel in the image. Two completely different colors can be of similar index values in a palette. Hence, it is always a challenging task to compress a color-indexed image as the compression must be lossless and predictive coding techniques are generally not effective to predict an index based on the spatial correlation of the index map. Palette reordering is a remedial process aiming at finding a permutation of the color palette to make the resulting color index map more suitable for predictive coding. Conventional palette reordering methods generally reorder palette colors to form a static palette whose index assignment is common to all pixels for making the reordering transparent to the decoder. In this thesis, an adaptive palette reordering method is proposed. Unlike those conventional palette reordering methods, this method adaptively reorders the palette to make the index assignment pixel-dependent. By so doing, the reordering is no longer transparent to the decoder. However, the resultant index map of the original color-indexed image can be of much lower zero-order entropy, smaller index variance and less spatial correlation, which makes the index map much easier to be encoded efficiently with a typical lossless codec such as JPEG-LS. Various lossless coding algorithms for color-indexed images can then be developed based on the proposed adaptive palette reordering algorithm. Simulation results show that their coding performance is better than state-of-art lossless compression algorithms including those are not based on palette reordering technique. In particular, when an index map was separated into binary bit planes with our suggested approach and then encoded with a context-based binary arithmetic coding scheme, an average compression ratio of 2.44:1 could be achieved. Index prediction, color reordering and DF-Table merging are some of the key functional components carried out in the proposed adaptive palette reordering algorithms. In practice, each one of them can be realized in different ways. Some of their realizations were evaluated and the result is also reported in this thesis.