Transform domain motion estimation without macroblock-based repetitive padding for MPEG-4 video

In order to perform motion estimation and compensation (ME/MC) of an arbitrarily shaped video object plane (VOP), macroblock-based repetitive padding is applied to the blocks on the borders of the reference VOP in MPEG-4 video. However the padding undesirably increases the overall complexity and data flow. In this paper, we propose transform domain motion estimation without repetitive padding for MPEG-4 video. It directly extracts motion vectors out of the two consecutive VOPs of the moving object. The resulted approach has low computational complexity, O(N²) compared to O(N⁴) for the modified block (polygon) matching motion estimation (MBKM-ME) in MPEG-4 video verification model (VM). Most importantly, the DCT-based nature enables the combination of both DCT and motion estimation units, thus the simplification of the heavily loaded feedback loop in MPEG-4 video coder design. It leads to a low-complexity, high-throughput fully DCT-based video coder. Moreover the proposed algorithm has inherently massive parallel operations and local connections suitable for VLSI implementation. Simulation demonstrates comparable performance of our transform domain approach and MBKM-ME in MPEG-4 video VM in terms of prediction errors