Flaw characterization using Mumford-Shah regularization for deconvolution

The ability to accurately detect flawed regions and estimate flaw sizes and shapes is important for practical NDE techniques. Most probes used in practice are not point probes, and the resulting measurement of a flaw tends to be "blurred", causing the footprint (or extent) of the flaw to appear to be larger and resulting in an overestimate of the flaw size. Image deconvolution methods, that model the measurement process as the convolution of the true flaw footprint with a probe blur kernel, have been used to determine the "true" flaw footprint by eliminating the effect of the blur kernel. In this paper, we examine a deconvolution technique based on Mumford-Shah regularization that assumes partial knowledge of the blur kernel, and has the ability to segment out the flaw region while simultaneously deconvolving the image. Results of applying the technique for detection and flaw size estimation from eddy current measurements are presented.