Dynamic Link Matching between Feature Columns for Different Scale and Orientation

Object recognition in the presence of changing scale and orientation requires mechanisms to deal with the corresponding feature transformations. Using Gabor wavelets as example, we approach this problem in a correspondence-based setting. We present a mechanism for finding feature-to-feature matches between corresponding points in pairs of images taken at different scale and/or orientation (leaving out for the moment the problem of simultaneously finding point correspondences). The mechanism is based on a macro-columnar cortical model and dynamic links. We present tests of the ability of finding the correct feature transformation in spite of added noise.