Department of Materials Science and Engineering, Drexel University, Philadelphia, 19104Department of Mechanical Engineering and Mechanics, Drexel University, Philadelphia, 19104.Corresponding author.
Abstract:
The generalized Hough transform is a common technique for feature detection in image processing. In this paper, we develop a size invariant Hough framework for the detection of arbitrary shapes in three dimensional digital microstructure datasets. The Hough transform is efficiently implemented via kernel convolution with complex Hough filters, where shape is captured in the magnitude of the filter and scale in the complex phase. In this paper, we further generalize the concept of a Hough filter by encoding other parameters of interest (e.g. orientation of plate or fiber constituents) in the complex phase, broadening the applicability of Hough transform techniques. We demonstrate the application of these techniques to feature detection in micrographs (2-D) and three-dimensional (3-D) microstructure datasets, and explore their utility to the closely related applications of feature based image segmentation and calculation of 3-D microstructure metrics.
