Age-Invariant Face Recognition Using Coupled Similarity Reference Coding

摘要

This paper presents a novel coupled similarity reference coding model (CSRC) by integrating non-negative constrained reference coding with coupled similarity metric for age-invariant face recognition (AIFR). Enhanced non-negative constrained reference coding is first proposed to encode the features into a meaningful part-based age-invariant representation, and we use iterative method combined with alternating direction method of multipliers to solve the troublesome constrained optimization problem. Coupled similarity metric is then proposed to eliminate the effect of irrelevant reference images in reference coding. By introducing non-negative constraint and coupled similarity metric, CSRC can obtain more meaningful representations and effectively enhance the performance of AIFR. Extensive experiments are performed on the CACD and MORPH datasets, which show an improvement in our model over state-of-the-art methods. What is more, experiment using deep features is performed and reaches high recognition rate, which shows that our model can be combined with deep networks to achieve better results.