A Novel and Complete Framework for Face Recognition with Pose Variations Using a Single Image

Abstract

A novel and complete framework for face recognition with pose variations using only one image is proposed in this paper. Firstly, feature points on face images are located with view-based AAM (Active Appearance Model), based on which, alignment and normalization are operated on face images. Secondly, mapping from non-frontal images to frontal images is constructed based on the algorithm of linear regression and frontal images are obtained from images with different poses. Finally, genetic algorithm is used to determine parameters of SVM that is applied to classify the facial features after dimension reduction. Experiments based on face database of CAS-PEAL-R1 show that performance of our proposed framework is better than other approaches for face recognition with pose variations. Recognition rates for face images with rotation of 15 degree, 30 degree and 45 degree can reach 98%, 84% and 76% respectively.