基于轮廓特征的图像配准

提出了一种新的基于轮廓特征的图像配准算法,首先提取图像轮廓并计算每个轮廓点的法向角,然后对轮廓点法向角进行直方图统计,通过对两幅图像的轮廓点法向角直方图进行圆周相关计算便可快速估计出两幅图像所存在的旋转角度.由于旋转角度参数的求出通常可以大大简化配准变换模型中其他参数的估计,因而这种方式可以实现快速配准.这种利用轮廓点法向角来估计旋转角度的方式具有旋转、平移和尺度不变性,并对轮廓缺失以及存在噪声的情况均具有很高的鲁棒性,可广泛适用于存在闭合轮廓与开轮廓的各种情况.本文所提出的配准算法已成功应用于PCB缺陷检测中实现了参考图像轮廓和待检测图像轮廓的快速精确配准.

Contour Feature Based on Image Registration

A new image registration algorithm is presented based on contour-feature, which involves extracting the contour of the image, calculating the normal angle of each contour point and proceeding normal angle histogram statistics, then estimating quickly the rotation angle between two images by circular correlation of normal angle histograms. The estimation of rotation angle can simplify the estimation of other parameters and realize the quick registration. The proposed algorithm is invariant to image rotation, translation and scale, and robust to contour deletion and noise, which is extensive to be applicable to open contours and closed contours. Moreover, the proposed algorithm has been successfully applied in Printed Circuit Board (PCB) defect inspection to realize quick and exact registration of reference image contour and the testing image contour.