Segmentation of weld defects in radiographs using rank-levelling and background subtraction

Abstract

Segmentation of defects in a digitised weld radiograph is a precursor to automatic interpretation of the radiograph. Simple thresholding has been shown by several researchers in the past to be unsuitable for isolating defects from the weld bead due to the non-uniform background intensity. The background subtraction method (BSM) using polynomial background estimation was shown by others to be able to produce satisfactory results. However, the BSM using polynomial background estimation is unsuitable when the background is highly non-homogeneous with noise present in the image. In this research, an effective algorithm has been developed to segment the defects automatically from noisy weld radiographs having poor illumination using background subtraction and rank-levelling. The rank-levelling method was used to obtain the background image of the digitised weld radiograph. A comparative study on background estimation by the rank-levelling technique and polynomial surface fitting algorithm shows that background estimation using rank-levelling is able to isolate the defects better compared to polynomial surface fitting algorithm.