Two-dimensional exponential discriminant analysis for small sample size in face recognition

Appearance-based face recognition methods usually require converting 2D face images to 1D column vectors. Due to high dimensionality and few available samples, the performance becomes unsatisfactory. In this paper, we propose a novel technique called two-dimensional exponential discriminant analysis (2DEDA) which extracts important features for classification directly from 2D face images. 2DEDA is especially suitable when the available sample size is small. Experimental results on two publicly available datasets viz. AR and CMU-PIE demonstrate the efficacy of the proposed technique, outperforming two-dimensional principal component analysis (2DPCA) and two-dimensional linear discriminant analysis (2DLDA). The performance of the algorithms is evaluated in terms of average classification accuracy.