提出了一种基于图理论的非刚体形状匹配算法。该方法在每次迭代过程中,先利用形状上下文算法获得待匹配形状点集的初始匹配,然后利用图理论剔除误匹配点,并估算匹配点集之间的TPS(thin plate spline)变换参数,再利用这些参数使待匹配点集相互逼近,最终实现非刚体的形状匹配。实验结果表明该算法提高了匹配的精度。
An algorithm based on graph theory for non-rigid shape matching is presented. In each iteration, the initial match for two shape point sets to be matched is obtained using shape context. Then error matching points are eliminated by making use of graph theory. And the thin plate spline(TPS) transformation parameters are estimated, by which the shape point sets are set closer each other. Finally, the non-rigid shape matching points are obtained. Experimental results demonstrate that the algorithm can improve matching precision.