Multi-stage iterative FLD method for face recognition

New multi-stage iterative FLD method For face recognition is proposed in this paper. In face recognition Principle Component Analysis (PCA) and Fisher Linear Discriminator (FLD) are used for recognition. Both have drawbacks like, in conventional FLD the training set is not sufficient to build a reasonable FLD basis for face recognition because of the involvement of classification information in the design process. In comparison, for PCA the training data is consider to be adequate. In order to provide reasonable training data-set for FLD basis we propose a multi-stage iterative FLD process. In each stage we consider only one, the most dominating base and move from stage to stage to build rest of basis. In this way we mitigate the effect of small training set size on the recognition performance achievable by FLD basis. The experimental results demonstrates the superiority of our approach over the existing ones.