An increment coefficient method for face recognition

In the paper we present an increment coefficient method used in face recognition which is also a linear representation-based method. Different from traditional linear representation-based method, for every class, we develop a linear model representing a virtual sample as a linear representation of the class-specific training sample and the testing sample. In the model, the virtual sample is the mean value of the class-specific training samples, the testing sample can be considered as an increment and we define the coefficient associate with the testing sample as the increment coefficient. We employ the regularized least square method to solve the inverse problem and label the testing sample as the class which has the maximum increment coefficient. Experiments were made on the Yale and ORL face database. The performance of our method was demonstrated on two face databases and compared to the state-of-art with linear representation algorithms.