Two Dimensional Principle Component Analysis of Gabor features for face representation and recognition

In this paper, a new technique called two dimensional Gabor principle component analysis (2DGPCA) is derived and implemented for image representation and recognition. The 2DGPCA method addresses the problems of feature extraction, feature selection and classification. In this approach, the Gabor wavelets are used to extract facial features. The principle component analysis (PCA) is then applied directly on the Gabor transformed image matrices in order to remove redundant information and form an efficient representation more suitable for face recognition. During classification, the Euclidean classifier is explored for simplicity and robustness in the presence of facial variations. The justification behind combining the 2D Gabor wavelets and the PCA is that Gabor transformed face images contain spatial locality, scale and orientation. These images are robust to variations due to pose, expression and scale, thus, making them most suitable for face representation and recognition. The proposed 2DGPCA method was tested on face recognition using the ORL database, where the images vary in expression, pose, and scale. In particular, the 2DGPCA method achieves 97.5% face recognition accuracy when using feature matrices of size 40times5times1 incorporating 5 different scales and 8 orientations compare to 93.5% and 94.5% with feature matrices of size 56times8 and 56times6 for the 2DPCA and 2DFLD method.