Joint systems of features for correlation pattern recognition

The method of calculation of a similarity measure of input and reference images is proposed. It is based on using two feature systems. First system is a contour preparation with increased width of contour lines. The second one is a map of directions of brightness drops with low resolution. Both systems of features have high invariant but low discriminant properties, however joint usage of these systems allows to essentially increase invariance to aspect transformations inside one class, as well as discriminant properties among different classes. Comparison of offered method with traditional correlation approach was carried out by two criteria.