A Comparison of Surfacelet-Based Methods for Recognizing Linear Geometric Features in Material Microstructure

Integration of material information and mechanical properties with geometry enables many product development activities, including design, analysis, and manufacturing. To integrate material information into CAD systems, geometric features of material microstructure must be recognized and represented, which is the focus of this paper. Linear microstructure features, such as fibers or grain boundaries, can be found computationally from microstructure images using surfacelet based methods, which include the Radon or Radon-like transform followed by a wavelet transform. By finding peaks in the transform results, linear features can be recognized and characterized by length, orientation, and position. The challenge is that often a feature will be imprecisely represented in the transformed parameter space. In this paper, we investigate several variations of the surfacelet based feature recognition methods, including masks, clustering methods, and whether to recognize features in the Radon or wavelet transform. These variations will be investigated to identify their strengths and limitations on a metal alloy and reinforced polymer microstructures.