摘要
提出了一种结合形状上下文分析的Laplace谱匹配算法。工作主要侧重于如何提高Laplace谱匹配算法对点的位置随机抖动的鲁棒性。首先,使用Laplace矩阵的特征向量和特征值以及双随机矩阵的方法计算初始匹配概率。然后,借助于概率松弛算法,将用形状上下文表示的局部相似性融入Laplace谱匹配算法以优化谱匹配的结果。对真实和合成数据的实验表明该方法具有比较高的精度。
A Laplacian spectral method combined with shape context analysis was proposed for point pattern matching. This work mainly focused on the problem of how to render the Laplacian spectral method robust for random position jitter. Firstly, the initial correspondence probabilities were computed by using the eigenvectors and eigenvalues of the Laplacian matrix as well as the method of doubly stochastic matrix. Secondly, local similarity evaluated by shape context was embedded into the Laplacian spectral method to refine the results of spectral correspondence via a probabilistic relaxation approach. Experiments on both real-world and synthetic data demonstrate that the method possesses comparatively high accuracy.
出处
《系统仿真学报》
CAS
CSCD
北大核心
2009年第14期4345-4350,共6页
Journal of System Simulation
基金
国家自然科学基金项目(10601001
60772121)
安徽省自然科学基金项目(070412065)
安徽省教育厅自然科学研究项目(2008B024
2008B142)
安徽大学211工程学术创新团队资助
关键词
LAPLACE谱
点模式匹配
双随机矩阵
形状上下文
概率松弛
Laplacian spectrum
point pattern matching
doubly stochastic matrix
shape context
probabilistic relaxation
