A New Invariant to Illumination Feature Descriptor for Pattern Recognition

A new descriptor for describing features in gray-scale images that is invariant to nonuniform illumination is proposed. The suggested method for the feature descriptor design is based on a local energy model which is a biologically plausible model of the visual system. The algorithm for feature detection and construction of the descriptor uses the scale-space monogenic signal framework and a modified algorithm for calculation of the histogram of oriented gradients based on the phase congruence of the signals. The results of computer simulation show that the proposed descriptor provides excellent detection and matching of features at nonuniform illumination, noise, and minor geometric distortions in comparison with known descriptors.