A Regularized Margin Fisher Analysis Method for Face Recognition
Margin Fisher Analysis is a typical graph-based dimensionality reduction technique and has been successfully applied to face recognition. However, it always suffers from the over-fitting, noise, and singular matrix problems. Common preprocessing methods such as PCA lose certain discriminant information in data, which leads the poor classification rate. We propose a novel method called Regularized Margin Fisher Analysis, which decomposes the inter-class similarity matrix into three subspace: principal space, noise space and null space. Then, we regularize the three subspaces in different ways to deal with the noise and over-fitting problems. Moreover, we use twice standard eigendecompositions instead of single generalized eigendecomposition which avoids the singular matrix problem. The experiments on Extended YaleB, CMU PIE and FERET face databases demonstrates that the proposed method is effective and can improve the classification ability.
KeywordsFace recognition Graph embedding Dimensionality reduction Regularization Margin fisher analysis
This work is partially supported by the National Natural Science Foundation of China (61402310). Natural Science Foundation of Jiangsu Province of China (BK20141195).
- 1.Basri, R., Jacobs, D.W.: Lambertian reflectance and linear subspaces. In: Proceedings of the Eighth IEEE International Conference on Computer Vision, ICCV 2001, vol. 2, pp. 383–390 (2003)Google Scholar
- 3.Belkin, M., Niyogi, P.: Laplacian eigenmaps and spectral techniques for embedding and clustering. In: NIPS, vol. 14, pp. 585–591 (2001)Google Scholar
- 5.He, X., Cai, D., Yan, S., Zhang, H.J.: Neighborhood preserving embedding. In: Tenth IEEE International Conference on Computer Vision, ICCV 2005, vol. 2, pp. 1208–1213. IEEE (2005)Google Scholar
- 10.Niyogi, X.: Locality preserving projections. In: Neural Information Processing Systems, vol. 16, p. 153. MIT (2004)Google Scholar
- 13.Sim, T., Baker, S., Bsat, M.: The CMU pose, illumination, and expression (PIE) database. In: Proceedings of the IEEE International Conference on Automatic Face and Gesture Recognition, pp. 46–51 (2002)Google Scholar