Limin Wang

An Efficient Implementation of Motion Estimation with Weight Prediction for ITU-T H.264 | MPEG-4 AVC

ITU-T H.264 | MPEG-4 AVC includes weight prediction in inter prediction process to improve coding efficiency when encoding video with a fading scene. Weight prediction, however, increases complexity of motion estimation (ME) process significantly in terms of additional operations and extra memory requirement. AVC performs weight prediction by scaling pixel value of the reference block to match with the fading effect. To follow this approach in ME, scaling operation has to be done at every pixel of all search blocks or precalculated before encoding. This paper proposes an efficient implementation method of weight prediction by scaling pixel value of the current block with an inverse of the weight factor instead. The proposed method reduces both the number of operations and memory requirement while sacrificing a negligible loss due to rounding error.

An Efficient Implementation of Unrestricted Motion Compensation in Video Encoder

This paper proposes an implementation method to reduce complexity of motion estimation in the padded areas for unrestricted motion compensation. The proposed method avoids repeating the same calculation of motion estimation cost in the padded areas with identical search outcome as the conventional motion estimation. The savings realized in this method depends only on the block size, image size, and the search range. The proposed method can be applied to several video compression standards such as H.263, and H.264.

Implicit transform unit partitioning in HEVC

Variable TU size is one of the key contributors to significant coding improvements in HEVC over traditional video coding standards. Residual quadtree is employed in HEVC to allow flexibility in TU partitioning while minimizing overhead bits. Flexibility in TU size also increases complexity in encoder dramatically. This paper described implicit transform unit partitioning which is employed in HEVC to reduce complexity in encoder while maintaining coding performance. Simulation results are also reported in this paper to validate coding performance of implicit TU partitioning.

Joint block motion estimation in H.264/AVC

Traditionally, block-based motion estimation (ME) searches for the best motion vector (MV) for one block at a time based upon a pre-defined ME cost. The ME cost is often a function of the best MV and its predictor (MVP). In H.264/AVC, MVP of a block is calculated from MVs of its neighboring blocks in the past. Therefore, the ME decision of a current block is affected by the neighboring blocks in the past, and in turn, it also affects the ME results of the neighboring blocks in the future. In this paper, we propose a joint ME process that searches for the best MVs for a group of blocks, which takes the interdependence of MVs into account. The proposed method smoothes MV fields and improves coding efficiency.

Fast motion estimation algorithm using unequal search effort for H.264|MPEG-4 AVC encoder

H.264IMPEG-4 AVC allows multiple reference frames and multiple partition types to improve the coding performance at the expense of much higher computation complexity. This paper proposes a fast motion estimation (ME) algorithm to reduce search complexity for this demanding configuration. Our experimental results show that the partition type which gives the lowest matching cost for the reference frame closest to the current picture in temporal domain is likely to be the best partition type for all reference frames. Based on this finding, the proposed method performs a less complicated search effort for the other partition types on the remaining reference frames to obtain significant savings in search complexity. This approach can be easily modified to support any existing fast search algorithms. Simulation results show the proposed algorithm significantly reduces the overall motion search complexity and at the same time, maintains minimum coding performance degradation.