A DCT-based Multimanifold face recognition method using single sample per person

One of the major drawbacks of the appearance-based face recognition methods is that they fail to work for face recognition from single sample per person (SSPP). In this paper, a new face recognition method based on the discriminative Multimanifold analysis (DMMA) in DCT domain is proposed to address the SSPP problem. For this goal DMMA algorithm is introduced and then DMMA in DCT domain is proposed. Indeed, two dimensional DCT is used as an initial feature extraction step and transforms local image patches from spatial domain to DCT domain with an aim to reduce noises and consequently produce better performance. Further, we statistically prove that the DMMA can be directly implemented in DCT domain. The experiments on a well-known face database FERET demonstrate that our modified algorithm can maintain better performance in both expression and pose changes. KeywordsDiscrete Cosine Transform (DCT); face recognition; manifold learning; Single Sample Per Person (SSPP); frequency features;