Multi-step Radiographic Segmentation of Weld Defect Images

Abstract

Non-destructive testing plays an important role in the evaluation of material properties as well as in the testing of a manufacturing system. The main objective of the proposed methodology is to improve the visual appearance of low contrast radiographic weld defect images without much loss of useful information by deploying multi-step radiographic enhancement (MSRE) and a subsequent region growing segmentation. The MSRE algorithm constitutes three major steps, namely linear weighting (LW), anisotropic diffusion filtering (ADF), and fuzzy image enhancement (FIE). This MSRE-based region growing segmentation methodology has addressed serious issues like amplification of noise, under and over enhancement, loss of edges, and image blurring. Further, the drawbacks of using single-step enhancement algorithms are overcome while preserving more useful image edges and details. This accounts for improving the segmentation accuracy of defective weld regions in non-destructive testing and evaluation (NDT&E). Following the three-step enhancement process, a region growing segmentation is performed on enhanced weld defect images to segment out the region-of-interest (ROI) defective regions. This multi-step segmentation methodology is tested on sample images from GDX-ray weld images database and the results are encouraging.