Adaptive arithmetic coding for image prediction errors

This paper presents adaptive arithmetic coding of prediction errors in lossless image compression. Generally, a probability distribution of the errors forms Laplacian distribution with zero mean, but the variance σ of the distribution may take different value at each local area in the image. The proposed encoder estimates the variance σ at every pixel to update the probability table. First, at a target pixel, the variance σ that maximizes the posterior probabilities of neighboring errors is calculated. Next, the error at the target pixel is encoded by arithmetic coding based on probability distribution with the variance σ. Since this method calculates the probabilities from fewer neighboring errors, they respond to the rapid changes of image characteristic in narrow area. In this paper, the proposed method is compared with Lempel-Ziv, Huffman, static/adaptive arithmetic coding and JPEG arithmetic coding, and then compression ratios are discussed. On an average, it generates 5% smaller size of compressed data than the adaptive arithmetic method by JPEG.