Recognition Using Region Correspondences

• Authors: Ronen Basri;David W. Jacobs
• Venue: International Journal of Computer Vision
• Year: 1997

Recognition systems attempt to recover information about theidentity of observed objects and their location in the environment. Afundamental problem in recognition is pose estimation.This is the problem of using a correspondence between some portions of anobject model and some portions of an image to determine whether the imagecontains an instance of the object, and, in case it does, to determine thetransformation that relates the model to the image. The current approachesto this problem are divided into methods that use “global”properties of the object (e.g., centroid and moments of inertia) and methodsthat use “local” properties of the object (e.g., corners andline segments). Global properties are sensitive to occlusion and,specifically, to self occlusion. Local properties are difficult to locatereliably, and their matching involves intensive computation.We present a novel method for recognition that uses region information.In our approach the model and the image are divided into regions. Given amatch between subsets of regions (without any explicit correspondencebetween different pieces of the regions) the alignment transformation iscomputed. The method applies to planar objects under similarity, affine, andprojective transformations and to projections of 3-D objects undergoingaffine and projective transformations. The new approach combines many of theadvantages of the previous two approaches, while avoiding some of theirpitfalls. Like the global methods, our approach makes use of regioninformation that reflects the true shape of the object. But like localmethods, our approach can handle occlusion.