Enhancements by weighted feature fusion, selection and active shape model for frontal and pose variation face recognition

Abstract

Face recognition is one of the most active areas of research in computer vision because of its wide range of possible applications in person identification, access control, human computer interfaces, and video search, among many others. Face identification is a one-to-n matching problem where a captured face is compared to n samples in a database. In this work a new method for robust face recognition is proposed. The methodology is divided in two parts, the first one focuses in face recognition robust to illumination, expression and small age variation and the second part focuses in pose variation. The proposed algorithm is based on Gabor features; which have been widely studied in face identification because of their good results and robustness. In the first part, a new method for face identification is proposed that combines local normalization for an illumination compensation stage, entropy-like weighted Gabor features for a feature extraction stage, and improvements in the Borda count classification through a threshold to eliminate low-score Gabor jets from the voting process. The FERET, AR, and FRGC 2.0 databases were used to test and compare the proposed method results with those previously published. Results on these databases show significant improvements relative to previously published results, reaching the best performance on the FERET and AR databases. Our proposed method also showed significant robustness to slight pose variations. The method was tested assuming noisy eye detection to check its robustness to inexact face alignment. Results show that the proposed method is robust to errors of up to three pixels in eye detection. However, face identification is strongly affected when the test images are very different from those of the gallery, as is the case in varying face pose. The second part of this work proposes a new 2D Gabor-based method which modifies the grid from which the Gabor features are extracted using a mesh to model face deformations produced by varying pose. Also, a statistical model of the Borda count scores computed by using the Gabor features is used to improve recognition performance across pose. The method was tested on the FERET and CMU-PIE databases, and the performance improvement provided by each block was assessed. The proposed method achieved the highest classification accuracy ever published on the FERET database with 2D face recognition methods. The performance obtained in the CMU-PIE database is among those obtained by the best published methods. Extensive experimental results are provided for different combinations of the proposed method, including results with two poses enrolled as a gallery.