Face Recognition Using Local Variation

In this paper, a novel approach, namely local variation projection (LVP), is presented for face recognition. LVP defines an adjacency graph to model the variation among nearby face images, which includes the within-class variation and between-class variation, also called margin. In order to better detect the discriminant structure, we assign a small weight to the variation among nearby face images from the same class. Based on this content, a concise feature extraction criterion is built for dimensionality reduction. Experiments indicate the effectiveness of our proposed approach.