Abstract

In this paper, we propose a new approach to overcome  the representation and matching problems in age invariant  face recognition. First, a new maximum entropy feature  descriptor (MEFD) is developed that encodes the  microstructure of facial images into a set of discrete codes  in terms of maximum entropy. By densely sampling the  encoded face image, sufficient discriminatory and  expressive information can be extracted for further  analysis. A new matching method is also developed, called  identity factor analysis (IFA), to estimate the probability  that two faces have the same underlying identity. The  effectiveness of the framework is confirmed by extensive  experimentation on two face aging datasets, MORPH (the  largest public-domain face aging dataset) and FGNET. We  also conduct experiments on the famous LFW dataset to  demonstrate the excellent generalizability of our new  approach.