提出了一种基于最小生成树与概率松弛结合的谱匹配算法。该算法分别对给定的两个待匹配的特征点集构建最小生成树,通过最小生成树构造Laplace矩阵,由奇异值分解该矩阵得到的特征值和特征向量,计算出特征点匹配的初始概率,利用概率松弛迭代法,获得最终匹配结果。用大量的真实序列图像进行比较实验,结果验证了该算法的有效性和准确性。
An algorithm of spectral correspondence based on minimum spanning tree (MST) combined with probabilistic relaxation in order to obtain accurate image matching is presented. Firstly, MST is constructed according to the feature points of two related images respectively. Secondly, the corresponding Laplacian matrixs for minimum spanning tree is constructed respectively. The original probability of point correspondence is gained by using the results of the SVD decomposition. Finally, the final matching results are acquired by using the method of probabilistic relaxation. The massive comparable experiments results show that the proposed method has the validity and accuracy.