基于集成人脸对距离学习的跨年龄人脸验证

摘要

针对不同年龄跨度下人脸对差异的不同,文中提出基于集成人脸对距离学习(EFPML)的跨年龄人脸验证方法.对不同年龄跨度的人脸对分别学习距离度量,然后使用集成方法对人脸对进行重表示,使人脸对重表示更具有判别性,并且可以扩充有限的跨年龄数据集.在公开的跨年龄人脸数据库FG-NET和CACD上的实验表明,文中方法可以有效减少年龄带来的影响,提高验证性能.%Aiming at the variations of face pairs caused by different age gaps, an ensemble face pairs distance metric learning method( EFPML) is proposed for cross-age face verification. Firstly, the whole dataset is divided into several subsets with different age gaps. Then, a distance metric is learned for each subset. Finally, all face pairs are re-represented for many times via learnt distance metrics, the new representations are more distinguishable and the limited cross-age face data are expanded. To evaluate the proposed method, a series of experiments are conducted on two real-world cross age datasets, FG-NET and CACD. The results show that EFPML consistently outperforms the state-of-the-art methods and it has ability to reduce the effect of aging and improve verification performance.