Abstract

In order to improve the recognition rate in single training sample, a face recognition algorithm based on invariant features of trace transform (IFT) is developed. In feature extraction, with application of Scharr operator and scale-adapted Laplacian of Gaussian (LoG) & Harris corner detector, keypoints are detected. Proper functionals are chosen to do trace transform in the circular region around each keypoint so as to obtain feature descriptors which are invariant to rotation and scaling. In feature matching, a coarse-to-fine matching strategy is executed by using the descriptors' feature vector and coordinate value. The identity of test image is obtained by comparing the matching score, which avoids the problem of parameters selection. Results show that it successfully improves the recognition rate under variations of illumination, occlusion, pose and expression. It also effectively reduces computational complexity and running time.

Paper Details