Near-lossless image compression: a key to high-quality data distribution

This paper promotes the use of near-lossless image compression and describes two DPCM schemes suitable for this purpose. The former is causal and is based on a classified linear-regression prediction followed by context-based arithmetic coding of the outcome residuals. It provides impressive performances, both with and without loss, especially on medical images. Coding time are affordable thanks to fast convergence of training. Decoding is always performed in real time. The latter is a noncausal DPCM and relies on a modified Laplacian pyramid in which feedback of quantization errors is introduced in order to upper bound reconstruction errors. Although the predictive method is superior for medium and high rates, the pyramid encoder wins for low rates and allows to encode and decode both in real time. Comparisons with block-DCT JPEG show that the proposed schemes are more than competitive also in terms of rate distortion.