Image registration and object recognition by affine invariant matching

This paper is concerned with the problem of feature point registration and scene recognition from images under weak perspective transformations, which can be approximated by affine transformations, under possible occlusions of the objects. The image can be represented as a set of feature points extracted from the outside contour of the objects, or the convex hull of scattered feature points. A second order Hopfield network model is constructed to establish the correspondence between the feature points of a test image with a reference image in order to recover the affine transformation between them. The differences of triangle area ratios between every 4 pair of points are used as compatibility constraints embedded in the energy function. The method has been tested on a series of real images and performs very well.