Sparse Representation Face Recognition Method with Fidelity Expressed as L2 Norm of Residual
In face recognition using sparse representation methods, the fidelity of sparse representation is typically expressed as the L2 norm of the residual. Maximum likelihood estimation theory demonstrates that this formulation requires residuals to follow a Gaussian distribution, an assumption that often fails in practical scenarios, particularly when test images contain abnormal pixels from noise, occlusion, or disguise. This limitation reduces the robustness of traditional sparse representation models built on conventional fidelity expressions. The maximum likelihood sparse representation recognition model addresses this by reformulating the fidelity expression as a maximum likelihood distribution function for residuals, transforming the maximum likelihood problem into a weighted optimization framework with enhanced robustness against abnormal pixels.