Face recognition by combining cauchy estimator and discriminant analysis

摘要

Face recognition is a crucial part of object recognition in robot research area, which the scholarly community has shown an intensive attention in the past few years. However, face recognition is still a difficult task since face images are easily confused by changes of the conditions, such as illumination, the different expression, or glasses. The goal of this paper is to address the problem when there are affected images in the dataset. Based on the Cauchy estimator theory and patch alignment framework (PAF), we proposed a dimensional reduction algorithm termed Cauchy estimator discriminant analysis (CEDA) for face recognition. CEDA not only preserves geometry structure of the input samples but also decrease the errors caused by confused samples. Extensive experiments were conducted on the UMIST dataset and demonstrated robustness and effectiveness of the proposed CEDA.