Invisible modification of the palette color image enhancing lossless compression

Abstract

We address the problem of pseudocolor image compression. Image values represent indices into a look up table (palette). Due to quantization, the neighbouring pixel values (indices) change too much. This deteriorates performance of both lossless and lossy image compression methods. We suggest a preprocessing phase that (a) analyses statistics of the adjacency relations of index values, (b) perforins palette optimization, and (c) permutes indices to palette to achieve more smooth image. The smoother image causes that the lossless imago compression methods yield less output data. The task to optimally permute palette indices is a NP complete combinatorial optimization. Instead of chocking all possibilities, we suggest a reasonable initial guess and a fast suboptimal hill climbing optimization. The proposed permutation of indices should enhance performance of most lossless compression method used after it. To our knowledge, the proposed reordering followed by our own nonlinear compression technique [HF97b, HF97a] yields the best compression. Experiments with various images show that the indices reordering provides data savings from 10% to 50%.