Research On Multi-modal Image Registration In PCB Infrared Fault Detection System

In PCB infrared fault detection system, seeking temperature fault point and locating PCBfault points fast is realized by the contrast of PCB chips infrared image, so the fast and accurate multi-modal image registration with robustness is necessary. Then the study of multi-modal image registration has great economic and practical significance.This paper especially introduces basic principles and relevant concepts of mutual information. On the basis, three different types of mutual information-- standard mutual information, gradient mutual information and regional mutual information is extended. After comparing the deviation, offset error and registration efficiency in registration result, the advantages and disadvantages of the three different types of mutual information are achieved.For the disadvantages of heavy computation and lengthy operation time in the registration technology based on mutual information, this paper resents a novel registration algorithm combining mutual information and image feature. Firstly, the chip edge in the infrared images and optical images is extracted by a model based on phase coherence. Next,the enclosed contour is formed by clipping and repairing using close value technology and ROI in infrared images and optical images is extracted by region growing algorithm in enclosed contour. Finally, the image registration experiment is performed by mutual information algorithm based on ROI. The experimental results demonstrate that on the premise of ensuring the registration accuracy, the algorithm improve the efficiency of the registration greatly, which has certain practicality and robustness.