Flaw detection in radiographic weldment images using morphological watershed segmentation technique

In this paper, the concept of application of morphological multistage watershed segmentation for detection of flaws in radiographic weld images is discussed. It is simple and intuitive and always produces a complete division of the image. The multistage watershed segmentation used here reduces the problem of over segmentation besides generating boundaries with very less deviation from their original position. Two-stage water segmentation is implemented here. At the first stage, watershed transform is applied to an X-ray image and the resultant mosaic image pattern is further thresholded by Otsu's thresholding method and converted into the binary image. Then, morphology and top-hat transformation is applied on binary image to separate partially overlapping objects. Euclidean distance map is calculated for each basin to label resultant segments uniquely and to separate ridges. This follows the second stage of watershed segmentation to obtain better-defined boundaries while removing over-segmented regions. Watershed segmentation algorithm has been able to detect flaws like slag inclusions and wormholes-type weld flaws. It shows all defects with reasonable accuracy having close contours. Similarly, small cavities are also highlighted successfully.