Representation and classification of iris textures based on diagonal linear discriminant analysis

Subspace methods are frequently used in pattern recognition problems aiming to reduce space dimension by determining its projection vectors. This paper presents subspace methods for feature extraction in an iris image called two-dimensional linear discriminant analysis (2DLDA), diagonal linear discriminant analysis (DiaLDA) and their combination (DiaLDA+2DLDA). The methods were applied in an UBIRIS image database, and the experimental results showed that DiaLDA+2DLDA overcame the 2DLDA method in recognition accuracy. Both methods are powerful in terms of dimension reduction and class discrimination.