Efficient Image Alignment Scheme for Automated Optical Inspection of Printed Circuit Boards

This paper investigates the application of automated optical inspection (ADI) in determining missing and incorrect components in printed circuit boards (PCBs) during the production line. Instead of employing computationally intensive CNN models, we adopt a simple image registration and matching based scheme. Despite of its simplicity, precise image alignment with the reference image of a flawless PCB is crucial to the inspection accuracy. Image alignment should overcome the impacts of camera shooting angles and the positional and angular deviations of the captured PCB image. A two-stage approach consisting of a coarse positioning and a fine calibration is developed. In stage 1, the reference image is first partitioned into inspection blocks and a Normalized Cross-Correlation (NCC) scheme is used to find the most similar blocks in the target image. In stage 2, feature points extraction and matching are performed under an iterative RANSAC (Random Sample Consensus) framework. Finally, a homography matrix is derived for fine calibration. Experimental results show that the alignment can be completed in 118.6ms and the average alignment error is less than 5 pixels.