Unified Framework For Classifying Facial Images Based On Facial Attribute- Specific Subspaces And Minimum Reconstruction Error

In this paper, a unified framework for classifying facial attributes is presented. Facial Attribute-Specific Subspace (FASS) is firstly proposed to represent each specific facial attribute. Then a framework is provided to classify facial images based on FASS and the Minimum Reconstruction Error (MRE) rule. The proposed framework is motivated by, but essentially different from the conventional Eigenface based methods, since, in our framework, similarity is measured by the reconstruction error. To evaluate the performance of the proposed method, it is applied to several face perception applications, such as face recognition, expression analysis, gender discriminating, and glasses detection. Extensive experiments in several face databases have demonstrated the impressive effectiveness and excellent robustness of the proposed framework against appearance variance due to changeable imaging conditions.