一种适合于低代价、低功耗VLSI实现的块匹配运动估值算法

A Block-Matching Motion Estimation Algorithm For Low Cost and Low-Power VLSI Implementation

本文给出了一种用于块匹配运动估值的改进的多分辨率望远镜搜索（ＭＲＴｌｃＳ）算法。它以望远镜的逆向搜索取代了传统的望远镜搜索，这一改进有效地降低了ＶＬＳＩ实现时对片上存储器容量和带宽的要求。此外本文还采用运动跟踪和自适应搜索窗技术来减小运动估值的计算复杂性。适合于低代价、低功耗ＶＬＳＩ实现是新算法的显著特点。模拟结果表明新算法要求的平均运算量仅为ＭＲＴｌｃＳ Ｈ法的３０％左右，而仍然可以得到相似的视频解码图质量。本文也给出了新算法和 ＭＲＴｌｃＳ算法用于 ＶＬＳｌ实现时的硬件代价和功耗比较。

This paper presents an improved multi-resolution telescopic search (MRTlcS) algorithm for the block-matching motion estimation. A novel inverse telescopic search is substituted for the conventional telescopic search. reducing the on-chip memory size and memory bandwidth when VLSI implementation. In addition, strategies of motion track and adaptive search window are applied to reduce the computational complexity of motion estimation. Simulation results show that, compared with the MRTlcS algorithm, the proposed algorithm reduces the computational load to only 30% while keeping almost the same image quality. Comparisons on hardware cost and power consumption of the VLSI implcmentations using the two algorithms arc also given in the paper.