Inter Frame Coding with Template Matching Averaging

Abstract

A template matching prediction based on a group of reconstructed pixels surrounding a target block enables prediction of pixels in the target block without motion information. The predictor of a target block is produced by minimizing the matching error of the template. Due to the freedom possessed by the template, the residuals of a target block may become large in flat regions. Our previous paper proposed to predictively encode the decimated version of a target block in flat regions to suppress the prediction errors. In this paper, the performance of template matching prediction is further improved. Multiple candidates are created by template matching at decoder. An average of the multiple candidates then forms the final predictor, which can reduce coding noise residing in the reference frames. Simulation results show that the proposed scheme improves coding efficiency of H.264 up to 7.9%.