摘要
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.
出处
《系统工程与电子技术》
EI
CSCD
北大核心
2010年第12期2691-2695,共5页
Systems Engineering and Electronics
基金
国家自然科学基金(60772121
10601001)
安徽省自然科学基金(070412065)
安徽省教育厅自然科学研究项目(kj2008b024)
安徽大学211工程学术创新团队资助课题
关键词
模式识别
谱匹配
松弛迭代
测地线-灰度直方图
LAPLACE谱
pattern recognition
spectral matching
iterative relaxation
geodesic-intensity histogram
Laplace spectral
