摘要： 提出2种结合颜色矢量的谱匹配算法。一种算法是从空间矢量关系的角度提取不受光源影响的图像颜色特征，结合图像特征点的几何特征，为待匹配的2幅图像分别构造亲近矩阵，通过对亲近矩阵进行奇异值分解构造一个反映特征点之间匹配程度的关系矩阵，从而获得匹配结果。另一种是将得到的匹配结果作为初始概率，通过双随机矩阵计算谱匹配概率矩阵，获得匹配的最终解。实验结果表明， 2种算法都具有较高的匹配精度。

关键词: 颜色矢量, 亲近矩阵, 奇异值分解, 谱匹配概率矩阵

Abstract: This paper proposes two spectral correspondence algorithms combined with color vector. One algorithm obtains image color features without the effect of different light conditions from the viewpoint of vector correlation in space. By combining with geometric feature of the feature points in images, the proximity matrixes of the two unmatched images are respectively defined. Through Singular Value Decomposition(SVD), a relation matrix that denotes the matching degree is constructed and the correspondence is obtained. Another algorithm computes the initial probability matrix from the correspondence, and acquires the final correspondence by using doubly stochastic matrix. Experimental results show that the algorithms are with comparatively high accuracy.

Key words: color vector, proximity matrix, Singular Value Decomposition(SVD), spectral correspondence probability matrix

中图分类号: 

• TP301.6