PCB Large Color Variation Image Registration with Local Optimization LoFTR

For the detection of Printed Circuit Board (PCB) defects with large color variations and large sizes, the traditional PCB image registration algorithm suffers from the problems of long time consumption and low accuracy, and the large color variation also interferes with the registration algorithm. We propose an image registration method for PCB with large color differences and large sizes to solve the problem that PCB image registration is easy to misalign. First, the large color variation problem of PCB images is corrected by the region-based color correction algorithm. Then, a local optimization feature matching algorithm is proposed for PCB image feature matching in response to the problem that the LoFTR algorithm loses the first window vector information. Finally, the MAGSAC++ algorithm is used to match PCB images. Experimental results show that the traditional feature matching algorithm is difficult to use for large-size PCB image registration, while the PCB registration method proposed in this paper has higher accuracy compared with the LoFTR algorithm, and further improves the registration accuracy by the region-based color correction algorithm.