结合测地线-灰度直方图的谱匹配算法

Spectral matching algorithm combined with geodesic-intensity histogram

针对单纯依赖奇异值分解的谱匹配方法的局限性,提出了一种结合测地线-灰度直方图和松弛迭代的Laplace谱匹配算法。首先,利用图像待匹配点集构造Laplace矩阵,通过对该Laplace矩阵进行奇异值分解,将得到的特征向量用于计算匹配概率;然后,引入具有局部特征的测地线-灰度直方图作为相容性约束,通过迭代的方式对匹配概率进行优化。实验结果表明,该算法实现了多特征、多算法的优势互补,提高了谱匹配算法的匹配精度和应用范围。

Aiming at the limitation of spectral matching by relying solely on singular value decomposition,a Laplace spectral matching algorithm combined with geodesic-intensity histogram（GIH） for point pattern matching is described.Firstly,the Laplace matrices are obtained from the point sets of the images.By using the eigenvectors of the matrices,the initial matching probabilities are computed.Then,the GIH with local similarity is introduced as a compatibility constraint.And the matching probabilities are refined via the iterative relaxation approach.Experimental results show the algorithm gets the complementation of multi-feature and multi-algorithm,and improves the matching precision and the application of the spectral matching method.