Efficient Block Size Selection in H.264 Video Coding Standard

This letter aims at selecting an efficient variable block size mode in H.264 video coding standard for better compression performance. The proposed scheme is based on 3D Recursive Search algorithm and takes into account the motion vector cost and previous frame information. The best mode for the current macroblock is obtained by analyzing the modes for a maximum of four macroblocks in the current and the previous frames. An improvement in the encoding time with negligible impact on the subjective and the quantitative performance has been achieved. INTRODUCTION: H.264 is a video compression standard being jointly developed by ITU-T Video Coding Experts Group and ISO/IEC Motion Picture Expert Group. The main goal of this standardization effort is enhanced compression performance and provision of a network-friendly packet-based video representation. H.264 uses a block based motion vector search algorithm. Several approaches have been presented in the literature to determine the best choice of motion vectors. Vaisey et al discussed techniques [1] in which the size of the block in motion estimation is varied according to the local detail of the image using quad-tree implementation. Kim and Lee described [2] a rate -distortion (RD) constrained approach for the hierarchical variable block size (VBS) motion estimation and displaced frame difference (DFD) coding. However, the previous published work does not take into consideration the block size information of the previously coded frame and the frame being coded in determining the variable block sizes. The frames in a video sequence are spatially and temporally related hence significant saving in motion estimation cost is possible by using variable block sizes and exploiting this correlation. This correspondence presents a new intelligent technique for the selection of the block size modes taking into consideration the variable block sizes calculated in the previous and current frames and their motion vector cost. The method presented here is comparable in coding performance to the reference scheme yet the computational complexity is reduced by over 30%.