Correcting image distortion in the X-ray digital tomosynthesis system for PCB solder joint inspection

Abstract

X-ray digital tomosynthesis (DT), which makes a cross-sectional image of 3D objects, has been researched and implemented in industrial applications nowadays, such as printed circuit board (PCB) inspection and inspection of electronic parts and other industrial parts/products. In this method, a cross-section image is obtained from a synthesis of more than two images projected from different views. However, distortion in X-ray images in practical imaging situation breaks the correspondences between those images and prevents us from acquiring accurate cross-section images. In this research, we propose a series of image correction method, which is composed of a neural network-based feature extraction for the distorted image and building a polynomial mapping function. The distorted raw image is sequentially corrected in terms of shape and intensity by using a reference pattern. To avoid corruption in feature extraction for the distorted image, an edge-filtered image is utilized rather than using a binarized one. Kohonen neural network is then employed to automatically group the edge points and localize the features, the pattern centers, without any pre-knowledge about the characteristics of the distortion. The proposed correction method is implemented to an actual DT system by carrying out a series of experiments on PCB. The results reveal the validity of the proposed image correction method and also verify the usefulness of the developed system for application of solder joint inspection.