Selection of thresholding methods for nondestructive testing applications

Abstract

In nondestructive testing (NDT) applications based on image analysis, image segmentation is the most important step in the extraction of defective regions of materials. An effective and very simple method of image segmentation, especially in NDT applications, is image thresholding. In an effort to present a quantitative evaluation of image thresholding methods NDT applications, 41 thresholding algorithms are compared. They are grouped in six categories based on the information they are exploiting, such as histogram shape, object attribute or clustering behavior, etc. Performance assessment is based on the weighted combination of four complementary objective metrics. Based on the results of such NDT images as defective thermal, ultrasonic, eddy current, etc., the thresholding algorithms that perform well over the majority of cases are established and a mixture thresholding scheme is proposed.