动态规划在运动图像分析中的应用  

Application of dynamic programming in dynamic image sequence analysis

作　　者：孙正[1] 郁道银[1] 

机构地区：[1]天津大学精密仪器与光电子工程学院,光电信息技术科学教育部重点实验室,天津300072

出　　处：《光电工程》2006年第3期32-35,61,共5页Opto-Electronic Engineering

基　　金：国家自然科学基金资助项目(30500129);高等学校博士学科点专项科研基金资助项目(20030056018);中国博士后科学基金资助项目(2004036361)

摘　　要：对运动图像序列中的目标进行运动跟踪和估计,可以分别采用变形模型和弹性配准的方法,两问题的求解均可归纳为成本函数的最优化问题。提出了应用动态规划算法求解离散成本函数最优化的方法,把问题分解成多个子问题分别求解,并存储子问题的解以避免重复计算。该方法不仅可保证解的全局最优性,而且计算量小,有利于实现实时处理。在对临床X射线冠状动脉造影图像序列的实验中得到了满意的结果,匹配大约2000个点的血管骨架只需要一秒钟左右的时间。对运动场已知的模拟图像的实验证明运动估计误差小于1个像素(1像素=0.3mm)。Tracking and estimating motion of the object from dynamic image sequences can be implemented based on deformable model and elastic registration technique, respectively. Both of them are actually optimization problems to minimize pre-defined cost functions. They can be solved by using dynamic programming （DP） to acquire a global optimal solution. The main idea of DP is to decompose the original problem into several sub-problems, solutions of which are stored to avoid redundancy. Its computation cost is rather low to realize real-time process. Experimental results of coronary arterial motion analysis from X-ray angiographic sequences are satisfactory. When matching about 2000 arterial skeleton points, we spent only l second. Results of the algorithm applied to synthetic images have shown that it is accurate, with a root mean square error less than 1 pixel （1 pixel=0.3mm）

关 键 词：动态规划 最优化 运动图像分析 运动跟踪与估计 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]