Automated grain boundary detection by CASRG

Constrained automated seeded region growing (CASRG) is an algorithm for automated grain boundary detection. It uses as input a single digitised microphotograph, such as ones obtained from a polarising microscope with an attached digital camera. In addition to this, it requires the user to click on the clasts within the microphotograph that the user wishes to obtain boundaries for. The algorithm requires no subsequent human input. The algorithm is based on the seeded region growing (SRG) algorithm of Adams and Bischof [Adams, R., Bischof, L., 1994. Seeded region growing. IEEE Transactions on Pattern Analysis Machine Intelligence 16, 641-647]. We have modified this algorithm to be guided by constraints and to adapt to the heterogeneity of colour information in the image. Imposition of these pre-determined additional conditions enables automated grain boundary detection without human intervention. The accuracy of CASRG has been validated through two benchmarking comparisons; one lithology with low tectonic strain and a second with high strain are used. The CASRG measurements are compared with those from hand drawn boundaries, which are used as a gold standard. Comparison is made using (a) a non-overlap statistic, (b) shape features, (c) strain estimates. In each case, the CASRG method compares very favourably with the gold standard. © 2006 Elsevier Ltd. All rights reserved.

Duke Scholars

Author Kingshuk Roy Choudhury Biostatistics & Bioinformatics, Division of Biostatistics