Block-level weighted interprediction with joint color components optimization for video coding

Modern video codecs employ two prediction methods: Inter-frame prediction (predicting a block in the current frame using a displaced block in a previously coded reference frame) and Intra-frame prediction (predicting a block using reconstructed causal pixels on the top and left of the block in the current frame). This paper proposes a block-level adaptive weighted prediction (BAWP) to improve inter-frame prediction efficiency by jointly optimizing the weighting parameters of Luma and Chroma components for video sequences with global and/or local illumination variations. For each color component, a weighted prediction model is derived at both encoder and decoder using causal neighboring samples, that minimize the squared errors between the reconstructed neighboring pixels in the current frame and the corresponding weighted predicted pixels from the reference frame. This method shows decent coding gains when implemented in the reference software of the next generation video codec of the Alliance for Open Media (AOMedia).