Symmetrical 2DLDA Using Different Measures in Face Recognition

Abstract

Facial symmetry can be regarded as a not absolute but useful and natural feature. In this paper, this symmetrical feature is applied to two-dimensional linear discriminant analysis (2DLDA) for face image feature extraction, furthermore, the distance measure (DM) and Frobenius-norm measure(FM) are also developed to classify faces. Symmetrical 2DLDA (S2DLDA) used pure statistical mathematical technique (just like 2DLDA), as well as the characters of face image (just like SLDA), to improve the recognition performance. The typical similarity measure used in 2DLDA is applied to S2DLDA, which is the sum of the Euclidean distance between two feature vectors in feature matrix, called DM. The similarity measure based on Frobenius-norm is also developed to classify face images for S2DLDA. To test their performance, experiments are performed on YALE and ORL face databases. The experimental results show that when DM is used, S2DLDA has the potential to outperform 2DLDA.