基于WαSH局部特征的立体影像匹配

Stereo images matching based on WαSH local features

针对传统方法对影像畸变敏感,难以用于具有较大变形的立体影像匹配的问题,提出了一种基于加权α形状(WαSH)局部区域特征的匹配方法.该方法采用WαSH特征检测算法提取图像的局部特征,并引入特征区域的二阶矩矩阵对椭圆拟合后的特征进行归一化,获得去除仿射变形的影像块;在归一化后的影像块上采用尺度不变特征变换(SIFT)描述算法提取特征描述向量,并基于最近邻与次近邻距离比率(NNDR)匹配测度进行特征匹配,确定同名特征.选用5组代表性影像进行实验.结果表明:提出的方法对发生较大视点变化、尺度+旋转变化、模糊及JPG压缩的影像均能获得正确率较高的匹配结果,与基于最大稳定极值区域(MSER)的匹配方法相比,匹配正确率及匹配效率更高.

Aiming at the problem that the traditional method is sensitive to image distortion and difficult to match the stereo images with large deformation,this paper presents a matching method based on the weightedα-shape（WαSH）local-invariant feature.In this method,the WαSH feature detection algorithm is used to extract the local features,and the second order moment matrices of the feature regions are introduced to normalize the feature regions to eliminate the affine distortion.The scale invariant feature transform（SIFT）description algorithm is employed on the normalized regions,and then the correspondences are determined based on the nearest neighbor distance ratio（NNDR）matching measure.Experimental results on five groups of representative images show that our method achieves high accuracy of matching result for stereo image pairs obtained on large scale viewpoint change,scale and rotation change,fuzzy change and JPG compression.Compared with the MSER-based matching method,the correct rate and efficiency of the proposed method are higher.