Face Recognition Using Diagonal 2D Linear Discriminant Analysis

Recently, two methods called 2D principle component analysis (2DPCA) and 2D linear discriminant analysis (2DLDA) were proposed to overcome the shortcomings in classical PCA and LDA such as time-consuming and singularity. They were used widely in image representation and recognition. A limitation of 2D methods is that, they only reflect the information between rows of images, while discarding columns information. More recently, a method called diagonal PCA (DiaPCA) was proposed to reserve the correlations between variations of rows and those of columns of images. DiaPCA seeks projection vectors from diagonal face images without image-to-vector transformation. However, DiaPCA does not involve any discrimination features, which are more than desirable in face recognition. In this paper, we propose a method named diagonal 2DLDA (Dia2DLDA) for face recognition. The proposed method projects face image samples by using an optimal projection vectors that derive from the diagonal representation of the image samples. Then, the projected image samples are employed to extract the discrimination features. The method compared with existing 2D schemes on two face databases. Results indicate that the performance of the Dia2DLDA overall better than other methods