Improving H.264 video coding through block oriented transforms

This paper introduces a pre-processing stage for the video coder H.264/MPEG4-AVC that takes advantage of the orientation of the blocks. Contrary to most of the proposed solutions in the literature, it is not the transform that adapts to the signal but the signal that is pre-processed to dasiafitpsila the transform. Residual blocks are straightened out to horizontal or vertical axes with the help of circular shifts applied at the pixel level. These shifts allow to simulate pseudo-rotations in the blocks. They are defined for all the partition sizes (16times16 to 4times4) and the two integer transforms of H.264 (8times8 and 4times4). Size and orientation for the blocks are evaluated with the help of a selection based on a rate-distortion criterion. This pre-processing stage applied on residual frames of a H.264 coder can improve its rate-distortion performances.