一种新的基于边缘拟合的图像配准方法

A novel image registration method based on edge fitting

为了克服图像配准中计算效率与稳定性的两难问题,提出一种新的基于边缘特征的图像自动配准算法。利用小波变换对图像进行多尺度边缘检测,提高边缘检测的稳定性;用直线拟合边缘,并根据几何特性、灰度特性为直线分配可信度;以直线对作为最小配准单位,利用马氏距离计算直线对之间的匹配程度;配准过程采用加权投票算法确定变换参数,每次投票根据直线对之间的匹配程度及各直线可信度拥有不同的权重。实验表明本文算法快速稳定,在没有任何先验知识的情况下,能够有效的配准存在相似变换关系(平移、旋转和尺度)的图像。

In order to tradeoff the computational efficiency and the stability in image registration,a novel image registration method based on edge features is proposed for image transform estimation.Image edges are extracted by using multi-scale edge detection based on wavelet transform to improve stability.Straight lines are used to fit these edges and every line is assigned a reliability value according to the geometric and gray characteristics.A pair of lines was taken for the basic matching unit.The similar degree between units from different images is computed by Mahalanobis distance method.The registration process is performed by a weighted voting algorithm,each voting holds different weights according to the similar degree between units and the reliability of elemental lines.The experimental result shows that this method is fast,stable and effective in dealing with similar transform(rotation,scale and translation)without any prior knowledge.