A variety of filtering algorithms are available, all of which exploit differences between defect and noise signals that originate in the intrinsically three-dimensional nature of the data. Methods include linear spatial decomposition using Gaussian filters, non-linear scale decomposition using rank-ordered filters and volumetric connectivity filtering, in which the filtered image only shows objects having a specified level of spatial connectivity between component voxels. The first two methods are developments of techniques that are well established in the field of image processing; connectivity filtering, however, is a relatively recent development taken from the more general field of morphological filters, and its use with ultrasound data is new.

The paper demonstrates the application of these methods to a variety of defect types in coarse-grained materials and shows that significant enhancement of image clarity, with suppression of noise-related features, can be achieved. A consensus-based approach is favoured, in which several filtering strategies are implemented in parallel, rather than just a single technique being universally applied. The paper also investigates the potential of these filters as the basis of an automated approach to data analysis and defect identification.