Weld defect detection in industrial radiography based on image segmentation

Radiographic testing is one of the most important non-destructive testing techniques for welding inspection. In this paper, a novel approach is proposed for welding defect detection on radiographic images based on a multi-scale segmentation strategy, where the initial partition is obtained using the minimal cut algorithm. The linear diffusion and an improved boundary trace method are then applied to implement multi-scale segmentation and extraction of the regions, which is followed by an energy-based evaluation model applied as stopping criteria to control the segmentation iteration. Therefore, the segmented sizes of defects obtained can be controlled by setting diffusion parameters according to the requirement of a special application. The proposed approach has been demonstrated by numerical experiments.