Local Lighting Invariant Features for Face Recognition

In this paper, a novel Local Lighting Invariant (LLI) model is proposed and applied to face recognition to extract lighting invariant features. In the LLI model, the TV-L-1 model based on nonlinear partial-differential equations (PDE's) is adopted to build a lighting invariant face image space first. In the built lighting invariant image space, the basic idea of orthogonal Laplacianface method is adopted to learn the local manifold structure of face samples. Experimental results on three famous lighting face databases (Yale Face Database B, extended Yale Face Database B and CMU PIE Database) confirm that the learned local manifold structure of faces by LLI has more discriminating power than orthogonal Laplacianfaces for face recognition.