A pattern recognition technique for the analysis of grain boundary structure by HREM

A pattern recognition technique for the detection of structural units in high resolution images of interfaces is described. The technique uses cross-correlation functions as a means of locating atomic patterns characteristic for an interface and as a measure of similarity between related units. Application is not limited to periodic, or even to planar interfaces. Characteristic structural units can be extracted from an experimental image and some important parameters such as mirror or mirror glide symmetry, and rigid body displacements can be determined without knowledge of the imaging parameters. The technique allows an image of a structural unit with reduced specimen noise to be obtained by averaging over several similar units, even if a boundary is not periodic and not planar.