奇异值控制的图像特征点匹配方法

Method of feature points matching in images controlled by singular values

图像特征点匹配是计算机视觉中需要解决的基本问题之一,准确地呈现匹配图像与待匹配图像特征点之间的一一对应是特征点匹配的最终目标。文中通过对特征点三角网格化,允许这些特征点适当偏离原先位置,同时通过线性分片映射Jocabian矩阵的奇异值来控制三角网格的变形,在这种约束下优化能量。实验结果表明,文中方法能够进一步提高匹配准确度。

Feature points matching of images whose ultimate goal is to present injective mapping accurately be- tween feature points in image and matched image, is considered as one of the fundamental problems in com- puter vision. In the paper, feature points are triangulated and allowed to deviate from initial position properly, at the same time, distortion of triangular mesh is controlled by the singular values of Jacobian matrix in piece- wise mappings. At last, energy is optimized under this constraint. It＇s shown by the results of experiment that this method can increase matching accuracy.