Face recognition using local invariant features

We propose automatic face recognition system based on multi-scale Harris corner detector, robust to the geometrical distortions of the images and to the changes of lighting conditions. A face region is segmented from given image using face segmentation algorithm. Segmented face region is represented by a set of interest points, detected by multi-scale Harris corner detector to achieve scale invariance. Each interest point is represented by a local feature vector, based on the Gabor filter, extracted at several scales and orientations to achieve rotation invariance. The illumination normalization is achieved by normalization of extracted Gabor feature vectors itself. The subspace principal component analysis method is used for further dimensionality reduction and variance maximization. The similarity between two faces is a measured by various distance metrics. Comparison of results with existing algorithms validates the usefulness of local invariant features for face recognition at less computational cost and less storage requirements.