Lossless coding of color quantized images based on pseudo distance

As pixels in quantized images such as GIF formatted images with 256 colors are usually represented by index numbers in a color palette, it is impossible to get high efficient compression by using conventional predictive coding to such images. In this paper, a novel predictive coding approach is proposed for images with quantized colors. In this approach, the prediction error is not an index number difference between an original color and its predicted color, but a value called as "pseudo distance" which is related to the Euclidian distance between these two colors in the 3-D color space. As the pseudo distance is small when the predicted color is perceptually close to the original color, the distribution of the pseudo distance is peak-like resulting in low entropy. Preliminary computer simulation results show that the proposed approach outperforms the index based linear prediction.