Micrograph analysis and image understanding

Abstract

Automating the analysis of rock section micrograph images is shown to present challenging problems analogous to those encountered in more general image understanding. Occlusion structures must be identified for a meaningful analysis of the images, and strategies for detecting these structures are presented. The method uses perceptual grouping principles. The identification of crystal faces on a cross-sectional contour is also shown to represent a perceptual problem, one which requires both 'local' and 'global' information about the contour structure. A successful multiresolution analysis scheme using these principles is described. The paper concludes by considering how the experience gained in developing principles of approach and implementing software for the analysis of micrographs can be transferred to the wider domain of generic object recognition and image understanding; several examples are given.