Abstract

We present a fast image comparison algorithm for handling variations in illumination and moderate amounts of deformation using an efficient geodesic framework. As the geodesic is the shortest path between two images on a mani- fold, it is a natural choice to use the length of the geodesic to determine the image similarity. Distances on the manifold are defined by a metric that is insensitive to changes in scene lighting. This metric is described in the wavelet domain where it is able to handle moderate amounts of deformation, and can be calculated ex- tremely fast (less than 3ms per image comparison). We demonstrate the similarity between our method and the illumination insensitivity achieved by the Gradient Direction. Strong results are presented on the AR Face Database.