Localized weighted prediction for video coding

The paper describes a new compression method for handling local brightness variations in video sequences. The method does not explicitly code and transmit the weighting parameters used to predict the current block. Instead, the weighting parameters are estimated by considering the correlation of the current block with its spatial neighbors and temporal predictors. Since the information for deriving these parameters is available in both encoder and decoder, no additional bits are required to be coded and transmitted. Compared to the weighted prediction method in the H.264 standard, our approach can further improve performance up to 15% in terms of bitrate reduction, while adding only minor complexity at the decoder.