Segmenting welding flaws of non-horizontal shape

Rapid detection of distortions formed in the welds of metal has high importance to pre -vent disasters. Radiography images of the weld have an unlimited number of small defects with diverse shapes. Visual inspection of the weld is a complicated, time-consuming task; and depends on the observers' experience. Many computer-aided detection techniques have emerged as an alter -native tool for segmenting the flaws of the weld, but none of them could segment alone all types of weld defects. In this paper, we attempt a possible solution and a novel image-based approach to solve the problem of weld defect detection using a data set of X-ray images. We aim to subtract the background and all horizontal defects and segment the non-horizontal flaws. For that, we apply some morphological operations and wiener filter to enhance the image quality; we use two novel filters to segment the non-horizontal defects; finally, we perform a post-processing operation. Our method depends on two convolution processes between the designed filters and the original image to achieve the segmentation process. We tested our approach on a universally available data-base of 68 images of the weld. Our method achieved high segmentation accuracy with zero errors. We highlighted the academic advancement of our technique by comparing its performance with other methods. Our efficient approach is effortless, applicable in practical segmentation processes of the defects of the weld. In the future, we will dedicate our work to the segmentation of horizontal defects by altering the shape of the two filters. Our method achieves satisfying accuracy; so, it is promising for weld defect detection. The particular contribution of our technique is that it can seg-ment alone all types of weld defects, contrary to traditional weld defect detection methods. The computing time is optimized compared to other algorithms. (C) 2021 THE AUTHORS. Published by Elsevier BV on behalf of Faculty of Engineering, Alexandria University.