Towards the Automatic Interpretation of Ultrasonic Non-Destructive Testing Data through the Application of Image-Thresholding and Region-Growing Segmentation

Abstract

To interpret ultrasonic non-destructive testing (NDT) B-scan images automatically, regions of interest (ROI) within the image must be identified, extracted, and interpreted. Image segmentation can be utilized for identifying constituent ROIs within images and their extraction for further analysis. Image segmentation is achieved either by applying edge detection transforms (EDTs), image thresholding (IT), or region growing (RG). Although EDTs can be applied without prior knowledge of the image, they are computationally expensive. By contrast, IT involves fewer computational processes but often requires some prior knowledge of the image and the selection of an appropriate threshold. However, B-scan images contain a background from which a threshold for segmentation can be established. Application of a suitable RG technique to the resultant image yields ROIs which can then be analysed and interpreted as being either defects or not. This paper outlines the development of a software system that utilizes IT and RG in order to extract ROIs within a B-scan image. Once extracted, these can be analysed within an automatic interpretation system for NDT data.