Abstract: Choosing the edges as the primitives for matching, during the edge detection the magnitude and direction of the edges and the Laplace values are figured out as three properties of the edge feature points. On the condition of Marr’s constraints for stereo match a 0-1 corresponding state space and a global energy function are built. With the state space transferring, this paper uses simulated annealing algorithm to minimize the energy function. When the energy function arrives the minimization, all the features in left image will find the corresponding features in the right images.

Key words: Stereo match, Energy function, Simulated annealing

摘要： 选择图像的边缘特征点作为匹配基元，以边缘梯度的大小、方向和拉普拉斯值作为特征点处的属性值，根据Marr的立体匹配约束条件，建立一个0-1对应状态空间和全局能量函数。最后用模拟退火算法，随着对状态空间的随机扰动，使能量函数达到全局最小，实现所有特征点的匹配。

关键词: 立体匹配, 能量函数, 模拟退火算法