How to extract a set of significant pixels in an image in order to represent the visual information, and to eventually derive 3-D information about the scene, is the subject of this paper. First we describe a relaxation method to detect contrast points in images, and then we propose an approach of maximum pattern resemblance based on the characteristics of these points to find point matches between pairs of images as well as the matches among images in a sequence. After a few iterations of the detection process, a `stable' pattern of discrete points is obtained and `good' corresponding points are then clustered successfully.
"Contrast pixel enhancement and matching in an image sequence by relaxation", Proc. SPIE 1613, Mobile Robots VI, (14 February 1992); doi: 10.1117/12.135190; https://doi.org/10.1117/12.135190