Addressing grain-matrix differentiation in sedimentary rock photomicrographs in the light of brightness perception modelling

Abstract

Brightness perception in real-life situations and subsequent image segmentation based on it is a complex phenomenon. The human visual system (HVS), apparently, does this effortlessly in the case of natural images. For specialized segmentation tasks, as in medical imaging, this is performed by the trained visual system of the specialist (radiologist, for instance, in medical imaging). In the present work, we shall concentrate on one such specialized task, viz. analysis of Photomicrograph of rock thin sections using petrological microscope, in the light of the HVS. For this, a new neural network model for the extended classical receptive field (ECRF) of Parvo (P) and Magno (M) cells in mid-level vision is elaborated at the outset. The model is based upon various well-known findings in neurophysiology, anatomy and psychophysics in HVS, especially related to the role parallel channels (P and M) in the central visual pathway. These two channels are represented by two different spatial filters that validate the reports of several psychophysical experiments on the direction of brightness induction. The mechanism of selecting the preferred channel for each of the stimuli consists of an algorithm that depends upon the output from an initial M channel filtering as captured in the visual cortex. We assume that the visual system of the geologist is training itself in the same way through such filtering processes in mid-level vision and identifying the important information in various situations in optical mineralogy and petrography. In the present work, the proposed model is applied in a simplified form on one such situation dealing with clast–matrix segregation from photomicrograph of sedimentary rocks, and is found to yield a promising result.