Lossless image compression using causal block matching and 3D collaborative filtering

Predictive coding has proven to be an effective method for lossless image compression. In predictive coding, untrans-mitted pixels are predicted based on the pixels already available at the decoder. Prediction errors are then compressed by entropy coding, and the original image can be reconstructed exactly at the decoder. More accurate prediction decreases the entropy of the prediction error, allowing for increased compression. Conventional image prediction methods rely on information from the immediate local neighborhood of each pixel. We introduce a novel predictor that leverages non-local structural similarities which have been shown to be effective in image denoising and deblurring applications. Experimental results show that the proposed method achieves state-of-the-art compression performance.