Mahesh Kumar Asati

Systematic Approach for Validation of X-Ray Automatic Defect Recognition Systems

With the advent of digital radiography, there has been a gradual shift from operators viewing images to find defects to totally automated defect recognition (ADR) systems. This has resulted in reduced operator subjectivity, reduced operator fatigue, and increased productivity. These automated defect recognition solutions are based on reference or non‐reference based approaches or a combination of both. There exists some amount of uncertainty or reluctance to accept automated systems in view of no systematic quantified metrics available on performance of these ADR systems in comparison to human operators. This paper describes the metrics that one could follow to quantify the performance of ADR systems such as detectability for different defect types and sizes, accuracy, false call rate, robustness to various noise levels etc., As it might be difficult to have images with defects of various sizes, shapes, contrast and noise levels, a methodology to generate images with simulated defects with variability’s i...