A New Approach for Defect Detection in X-ray CT Images

We introduce a novel method to automatically evaluate Xray computed tomography (CT) images for the purpose of detecting material defects by evaluating the significance of features extracted by first order derivative filters. We determine the noise characteristics of an image using robust estimation methods and compute the noise of the filtered image via error propagation. The significance of these features can then be evaluated based on the signal-to-noise ratio in the filtered image. The major benefit of that procedure is, that a sample-independent threshold on the signal-to-noise ratio can be chosen. The results are demonstrated on parts drawn from an industrial manufacturing line.