An Age Invariant Face Recognition Algorithm Based on Decorrelated Feature Decomposition

As people grow old, the intra-class difference caused by face aging is also more and more large, which has a great impact on the effect of face recognition. However, the identity feature obtained by the feature decomposition of the previous AIFR algorithms were always doped with age-dependent feature to a certain extent, which failed to further reduce the intra-class differences caused by age factors. To overcome this problem, a new algorithm is proposed. Firstly, in order to extract the initial facial feature of the face, the full pre-activation residual block is used to construct the backbone of the model, which effectively improves the generalization ability of the model; Secondly, in order to realize feature decomposition, age-dependent feature are extracted through two fully connected layers, and age-independent feature are obtained by subtracted age-dependent feature from initial facial feature. Finally, in order to reduce the intra-class differences caused by age factors, the Euclidean distance between age-dependent and age-independent feature is maximized to further reduce the correlation between them. A large number of evaluation experiments have been conducted on the public age-invariant face recognition (AIFR) datasets (Morph and CACD-VS) to verify the effectiveness of proposed method.