Hierarchical segmentation approach to detection of defects on welding radiographic images

Detection of defects on welding radiographic images has been a challenging task since radiographic films normally have deficient contrast caused by the inspection techniques. In this paper, a region-based hierarchical detection approach is proposed to remedy this problem that is based on multi-scales segmentation strategies. The proposed method consists of two major stages: initial partition and iterative refinement until all defects are found out. In which energy-based evaluation models are proposed as stopping criteria for the controlling of the defect scales segmented such that only the desired defects in specific scale are detected to meet the requirements of different applications. The proposed algorithm has been demonstrated by several numerical experiment results.