Face Oriented Discrimination Gabor Features

Gabor face representation has been getting popular in face recognition applications; however, it also suffers from the high dimensional data containing diverse redundancy and different random noise. This paper deals with analysis and selection of discriminating features in face recognition. First the feature differences between every two face images within a training data set are calculated and grouped into two categories: intra individual set and extra individual set. Then the rank of discriminating capabilities and redundancy between features can be estimated by analyzing the distributions of feature differences in intra set and extra set based on statistics and mutual information theory .First, an approach for ranking the discriminating abilities of Gabor features singly by mutual information analysis is proposed. In view of the widths of the Gabor filter convolutions at adjacent nodes exist considerable overlap, a simply sequential elimination redundancy approach based on joint mutual information is also proposed in this paper. The rank and the optimal feature combination selection with the proposed approach had been implemented on the training set of the CAS-PEAL- Rl database, and some testing experiments were implemented on probe sets of the CAS-PEAL-R1 database and FERET face database. The experiment results show that taking only about half features with top discriminating rank can achieve nearly the same recognition performance as taking full set features. A more sophisticated approach based on combining the rank information and elimination of redundancy by analysis joint of conditional mutual information can get even better performance yet only using one fifth of the full features.