Robust error correction with multi-model representation for face recognition

摘要

The present invention provides a face recognition method on a computing device, comprising: storing a plurality of training face images, each training face image corresponding to a face class; obtaining one or more face test samples; applying a representation model to represent the face test sample as a combination of the training face images and error terms, wherein a coefficient vector is corresponded to the training face images; estimating the coefficient vector and the error terms by solving a constrained optimization problem; computing a residual error for each face class, the residual error for a face class being an error between the face test sample and the face test sample's representation model represented by the training samples in the face class; classifying the face test sample by selecting the face class that yields the minimal residual error; and presenting the face class of the face test sample.

主权项

1. A face recognition method on a computing device, comprising:
storing a plurality of training face images, each training face image corresponding to a face class, each face class including one or more training face images; obtaining one or more face test samples; applying a representation model to represent the face test sample as a combination of the training face images and error terms, wherein a coefficient vector is corresponded to the training face images; estimating the coefficient vector and the error terms by solving a constrained optimization problem, including:
presenting a general framework by using a multi-model representation of error terms to formulate the constrained optimization problem,choosing one or more application-specific potential loss functions for characterizing the error terms,choosing a method for regularization, andsolving the constrained optimization problem,wherein each application-specific potential loss function is a summation of dual potential functions; computing a residual error for each face class, the residual error for a face class being an error between the face test sample and the face test sample's representation model represented by the training samples in the face class; classifying the face test sample by selecting the face class that yields the minimal residual error; and presenting the face class of the face test sample; wherein:
provided that that yεd denotes a face test sample and T=[Ti, . . . , Tc]εd×n denotes a face dictionary, a face dictionary being a matrix with a set of training samples of c subjects stacked in columns, Tiεd×ni denotes the ni set of samples of the ith subject, such that Σi ni=n, a denotes the coefficient vector, J(a) denotes a function of the coefficient vector a,the general framework to formulate the constrained optimization problem is defined to estimate the coefficient vector a that minimizes:
J(a)=Φk(y−Ta)+λθθ(a) wherein function Φk(x) represents an application-specific potential loss function, and function θ(•) defines the method of regularization.