Block motion estimation using luminance transform based matching

In this paper, a luminance transform based block motion estimation method is proposed to increase the prediction performance of the full search (FS) block matching algorithm. The most similar candidate for each block of the current frame is determined from the previous frame using the mean absolute difference (MAD) criterion. If the MAD value of the best matching candidate block is below a certain threshold this match is kept. Otherwise this match is regarded to be unsatisfactory, and a luminance transform based matching is performed. The luminance transform parameters are multiplexed by the encoder into the bit stream and used at the decoder in addition to the motion vectors for reconstruction. It is shown that this approach outperforms conventional MAD based FS (FS-MAD) motion estimation in terms of objective and visual performance.