A geometric invariant representation for the identification of corresponding points

The identification of the corresponding points between two recordings of a point set has been always an important problem in stereo vision and image registration applications. We propose a geometric invariant representation for each point in order to match the two point patterns and determine their corresponding points. Each point is represented by a feature vector computed in terms of the relative positions of a small number of neighborhood points with respect to the given point. The matching algorithm pairs each point of one set with the most similar one in the other set based on the similarity degree of their feature vectors. A consistent set of corresponding point pairs is therefore extracted independent of any geometrical transformation which may alter one point set with respect to the other one.