Detection of edge structures on surface of sedimentary grains acquired by electron microscope

This paper presents a method for edge detection on the surface of sedimentary grains that were acquired by an electron microscope. Local grain parts are described by textural co-occurrence features. Edges are then detected by classification of co-occurrence features corresponding to particular parts of image. For this classification, a logistic regression model is used. The precision and recall values of the cross-validated model are 82% and 77% respectively. Further, a measure that quantifies a maximal edge length detected on a grain is proposed. The purpose of this measure is to provide a high-level feature for comparing different grain sets. To evaluate a usability of the measure, the measure is computed for sets of grains of different geomorphological geneses and the differences are compared. Because the results showed a specific measure range for some geneses, the proposed edge detection method can be considered as useful for description of sedimentary grains.