A lossless image coder integrating predictors and block-adaptive prediction

This paper proposes a lossless image compression scheme integrating well-known predictors and Minimum Rate Predictor (MRP). Minimum Rate Predictor is considered as one of the most successful method in coding rates for lossless grayscale image compression so far. In the proposed method, the linear predictor is designed as the combination of causal neighbors together with well-known predictors (GAP, MED, and MMSE) to improve coding rates. To further reduce the residual entropy, we also redesign the calculation of context quantization and the disposition of neighboring pixels. The modifications made in our proposed method are crucial in enhancing the compression ratios. Experimental results demonstrate that the coding rates of the proposed method are lower than those of MRP and other state-of-the-art lossless coders among most of the test images. In addition, the residual entropy of the proposed scheme in the first iteration is lower than that of MRP and is relatively closer to the final residual entropy than that in MRP. This phenomenon will allow our proposed scheme to be terminated in less iterations while maintaining a relatively good compression performance.