基于局部轮廓曲线的图像配准方法

Image registration method based on local contour curve

图像配准是计算机视觉领域的基本问题.提出了一种基于局部轮廓曲线的图像配准方法.首先,利用图像的最小作用曲面提取从角点出发的局部轮廓曲线,并通过MLSD描述符描述轮廓曲线;其次,对曲线的MLSD特征向量进行傅里叶变换以便于使用轮盘匹配方法实现两幅图像中从角点出发的局部轮廓曲线的配准;然后运用最小二乘法迭代求解局部配准参数;最后,统计局部配准参数直方图获取全局配准参数.实验结果表明:此配准方法有较高精度.

Image registration is a basic problem in the field of computer vision.In this paper,an image registration method based on local contour curvesis presented.Firstly,the local contour curves starting from the corners are extractedby surface of minimal action of the images and described by MLSD descriptor.Secondly,the MLSD eigenvectors of the curves are transformed by Fourier transform so as to use the roulette matching method to achieve the registration of the local contour curves starting from the corner points in the two images.Next,the least square method is utilized to solve the local registration parameters iteratively.Finally,the local registration parameter histograms are calculated to obtain the global registration parameters.The experimental results show that the proposed registration method has high accuracy.