Reassembling 2DLDA for face recognition

Two dimensional linear discriminant analysis(2DLDA) provides a solution to the small sample size(S3) problem presented to the classical LDA. However, it takes each column of the image matrix, which only contains partial information of the whole human face, as an input vector. In this paper, a novel reassembling 2DLDA (R2DLDA) algorithm is proposed for face recognition. The new reassembling 2D sample, each column of which is consisted of a sub-image of original face image, is introduced. Then 2DLDA is applied for face recognition. Experimental results on the ORL face database show that the proposed R2DLDA algorithm is feasible and has higher recognition rates than original 2DLDA and Alternate-2DLDA algorithms.