Techniques for Automated Point Matching in Image Pairs

Dr. Stephen Dow

Department of Mathematical Sciences
University of Alabama in Huntsville

October 20, 2000

Abstract

Point matching is a basic operation required for various other higher level tasks in photogrammetry and computer vision. The term refers to identifying corresponding points in a pair of photographs, i.e. points which are the images of a common feature in the 3D world. A large literature now exists on various techniques for automatically generating such point correspondences. The techniques focus on (1) interest operators, such as corner detectors, used to extract points with good matching characteristics from one or both images, (2) operations such as correlation used to generate candidate matches, (3) algorithms for extracting a set of consistent point matches from a larger set of candidate matches, and (4) least squares refinement of matches to obtain subpixel accuracy. In this talk we will survey these techniques, briefly present some of the mathematics they involve, and illustrate them with computer demonstrations.