文中给出了一种用于实现多分辨率运动估值算法后阶段任务的改进的树结构。在一个简单的RISC类型核控制下,它能够完成整个运动估值过程中除粗分辨率精度运动矢量搜索之外的所有后阶段子任务。包括运动矢量优化(搜索)在内的多任务是通过二叉树最底层叶节点上的多功能处理单元和可以拆分成子树的加法树来实现的。此外,运算单元寄存器堆的设计使能在二维方向上复用图像数据,完全避免了同一类数据从存储器中重复读取,从而实现了最小的存储器访问带宽,并有助于减小存储功耗。文中給齣瞭一種用于實現多分辨率運動估值算法後階段任務的改進的樹結構。在一箇簡單的RISC類型覈控製下,它能夠完成整箇運動估值過程中除粗分辨率精度運動矢量搜索之外的所有後階段子任務。包括運動矢量優化(搜索)在內的多任務是通過二扠樹最底層葉節點上的多功能處理單元和可以拆分成子樹的加法樹來實現的。此外,運算單元寄存器堆的設計使能在二維方嚮上複用圖像數據,完全避免瞭同一類數據從存儲器中重複讀取,從而實現瞭最小的存儲器訪問帶寬,併有助于減小存儲功耗。 문중급출료일충용우실현다분변솔운동고치산법후계단임무적개진적수결구。재일개간단적RISC류형핵공제하,타능구완성정개운동고치과정중제조분변솔정도운동시량수색지외적소유후계단자임무。포괄운동시량우화(수색)재내적다임무시통과이차수최저층협절점상적다공능처리단원화가이탁분성자수적가법수래실현적。차외,운산단원기존기퇴적설계사능재이유방향상복용도상수거,완전피면료동일류수거종존저기중중복독취,종이실현료최소적존저기방문대관,병유조우감소존저공모。 An improved tree architecture is proposed for implementing thepost-stage tasks of the multi-resolution bloc k matching motion estimation algorithm. Based on the tree architecture, all th e tasks of motion estimation except the coarse search can be realized under the control of a simple RISC-like core. In the architecture, multi-function proc essing elements are arranged on the lowest layer of the binary tree and the adde r tree can be split into two sub-trees. Besides, register files are designed for each computing unit in order to reuse data in two-dimensional directions, thus fully avoiding reading the same data from memory repeatedly and achieving m inimum memory bandwidth. And the power dissipation by accessing memory is also reduced.