Robust local feature extraction algorithm with visual cortex for object recognition

A robust local feature extraction algorithm comprised of two key steps is proposed. The first step is the extraction of feature point candidates from a multi-scale fast Hessian detector at 0° and 45° based on Gabor rotation angles and a multi-scale Harris corner detector based on the Gabor phase. The second step is a MAX operation for selecting points from the outputs of the three feature point detectors used in the previous step. Performance analysis shows that the proposed algorithm significantly enhances the recognition rate and is better than SIFT and SURF.