Edge detection in petrographic images

Abstract

Summary The automatic detection of mineral grain boundaries in images obtained from a polarized‐light microscope requires special techniques. Observations in both plane‐ and cross‐polarized light may be necessary and the section must be rotated relative to the plane of polarization of the microscope to see all the grain boundaries. In computer‐based microscopy this can be accomplished by the sequential accumulation of individual images captured from one microscope field of view with different polarizer orientations. For real‐time implementation the sequential images are segmented individually by applying Canny's algorithm. A separable Gaussian mask is used for smoothing and a 3 times 3 convolution mask is used to generate 1‐pixel‐wide boundaries, which are located at the zero‐crossing of the second‐order derivative of the intensity gradient. The boundaries are extracted and accumulated in a composite image. The resulting composite image is a synoptic grain‐boundary image of the rock.