基于MSURF算法的PCB图像配准

A modified SURF algorithm for PCB image registration

针对PCB板缺陷检测中用传统SURF算法进行图像匹配精度不高的问题,提出一种生产工序中运动平台机械误差先验信息与SURF特征提取相结合的MSURF配准算法。通过提取计算SURF特征点,求出对应特征点对的距离;在PCB运动平台机械误差分析的基础上,依据先验阈值边界条件筛除异常匹配特征点对;求出两幅图像满足最小二乘拟合准则的映射关系,并将其用于图像配准。在机械运动误差0.05~0.10 mm范围内对42 mm×42 mm的PCB图像配准实验,结果表明:提出的图像配准方法速度快、精度高,适用于产线PCB缺陷检测。

Aiming at the low accuracy of the traditional SURF algorithm in image matching for PCB board defect detection,a modified SURF registration algorithm was proposed that combined the prior information of the mechanical error of the motion platform and the SURF feature extraction in the production process. Firstly,the distance between the corresponding feature point pairs was obtained by extracting and calculating the SURF feature points. Secondly,based on the mechanical error analysis of the PCB motion platform,the abnormal matching feature point pairs were screened according to prior threshold boundary conditions. Finally,the mapping relation of two images satisfying the least square fitting criterion was obtained and used for image registration. The results of the 42 mm×42 mm PCB image registration experiments conducted with the error range of mechanical motion controlled to be 0.05 ~ 0.10 mm indicate that the image registration method proposed is fast and accurate,thereby suitable for PCB defect detection in production.