CAJ | 학술논문

本文给出了一种用于MPEG2MP@ML视频编码器单片集成的低功耗运动估值器实现，它基于一种低复杂性的改进的多分辨率望远镜搜索算法，在粗分辨率运动矢量搜索过程中,采用以运动跟踪搜索和自适应望远镜搜索组成的两阶段可提前中断搜索来替换单一的传统望远镜搜索.模拟结果表明新算法的计算负载仅为传统算法的30％左右，却可以保持与之相同的视频解码图质量.本文采用心动阵列和改进的树结构阵列来分别加速运动矢量的粗分辨率搜索和优化搜索，并通过门控时钟和屏蔽操作数等方法来实现低功耗.
본문급출료일충용우MPEG2MP@ML시빈편마기단편집성적저공모운동고치기실현，타기우일충저복잡성적개진적다분변솔망원경수색산법，재조분변솔운동시량수색과정중,채용이운동근종수색화자괄응망원경수색조성적량계단가제전중단수색래체환단일적전통망원경수색.모의결과표명신산법적계산부재부위전통산법적30％좌우，각가이보지여지상동적시빈해마도질량.본문채용심동진렬화개진적수결구진렬래분별가속운동시량적조분변솔수색화우화수색，병통과문공시종화병폐조작수등방법래실현저공모.
An implementation of a low-power motion estimator dedicated toMPEG2 MP@ML is proposed.It adopts an improved multi-resolution telescopic search algorithm with low complexity.For the coarse search procedure,a new two-stage method,in which motion track search is conducted followed by adaptive telescopic search,is used instead of the unitary conventional telescopic search.In addition,the two-stage search can be early ended in accordance with the motion situation.Simulation results show that,compared with the conventional algorithm,the new algorithm reduces the computational load to only 30% while maintains same quality of the decoded pictures.To accelerate the two procedures of coarse search and motion vector refinement,a systolic array and an array of improved tree architecture are devised.Techniques such as clock gating and operand isolation are used to reduce the power consumption of the motion estimator.