Adaptive linear prediction for lossless video coding

Linear prediction is an efficient tool for lossless video coding; however data overfitting and context dilution often occur which reduce the prediction efficiency, especially in occlude and large motion image area. In this paper, we will adopt intra and inter two liner prediction modes to eliminate the redundancy among temporal and spatial neighbor pixels. The two prediction modes can be adaptively switched based on their prediction accuracy evaluation in predefined neighborhood region. Experiment results shows that our proposed method can improve prediction efficiency up to 0.27dB.