蝶形运动估计算法及其VLSI实现  

Butterfly-shaped Motion Estimation Algorithm and its VLSI Implementation

作　　者：田应洪[1] 何俊[1] 洪志良[1] 

机构地区：[1]复旦大学专用集成电路与系统国家重点实验室,上海201203

出　　处：《小型微型计算机系统》2008年第4期741-745,共5页Journal of Chinese Computer Systems

基　　金：华为科技基金项目(YJCB2005019BA)资助

摘　　要：视频技术发展要求更高速,更利于硬件实现的运动估计算法.提出了一种蝶形运动估计算法,该算法采用蝶形搜索模板、快速截止技术和运动向量预测技术.该算法较钻石搜索算法提速43.26%-80%,并且图像质量更好.同时,本文采用加法树和片内并行存储器,构建该算法的VLSI实现结构.通过两种数据映射方法(拉丁方映射和4×4块映射),该结构不但解决了快速搜索算法的数据不规则性难题,并且节省了带宽.当系统时钟为27MHz,数据总线为16位,外部存储器带宽要求仅为4.57Mbit/s.比较其它硬件实现结构,该结构采用了更少的处理单元数,更小的缓存单元,但却获得更快的速度和更高的灵活性.Development of video technology requires the fast and hardware-friendly motion estimation algorithms. A novel fast motion estimation algorithm with butterfly-shaped search pattern, halfway-stop technique, and motion vector prediction is proposed. This algorithm can achieve 43. 26-80 percent speedup than diamond search algorithm and the picture quality is better. And then a relative VLSI architecture with tree-structured adders and parallel internal memories is also proposed and analyzed. Two new access methods for parallel memories, Latin square map and 4 × 4 block map, are adopted in this VLSI architecture, which regularize the data and save the external memory bandwidth. Under the condition of the 27MHz system clock and the 16 bits data bus, the desired data rate of external memory is only 4. 57Mbit/s. Compared with other implementations with tree architecture, this implementation has superior performance on PE count, memory size, search speed and flexibility.

关 键 词：视频图像 运动估计 拉丁方 加法树 

分 类 号：TN919.8[电子电信—通信与信息系统]