Revisiting the JPEG-LS prediction scheme

Abstract

The authors investigate the prediction scheme of JPEG-LS, the latest JPEG standard for lossless/near lossless image compression. They show that it is not sufficient to consider only horizontal and vertical edges in constructing predictive values. As a result, they propose an additional diagonal edge detection scheme to achieve better prediction accuracy and hence provide potential for further improvement. Experiments show that, in terms of mean-square-error values, the proposed scheme outperforms the existing JPEG-LS prediction for all images tested, while the complexity of the overall algorithm is maintained at a similar level.