针对跨姿态人脸识别的度量学习方法

A Metric Learning Method for Pose-Invariant Face Recognition

近年来,由于深度学习技术的引入,人脸识别技术取得显著的发展。然而,当前的人脸识别模型在解决跨姿态人脸识别问题上效果仍然不理想。其中导致这一现象的主要原因是,目前用来训练人脸模型的数据集中姿态变化较少或者不均衡。针对跨姿态人脸识别问题,提出一种基于度量学习的方法 CPP Loss。该方法能够有效地利用训练集中有限的姿态变化,在基准模型上进一步提升其在跨姿态条件下的人脸识别准确率。

Recently, due to the emergence of deep learning, face recognition has made remarkable progress. However, many contemporary face recognition models still perform relatively poor in solving pose-invariant face recognition problem. A main reason is that the most datasets used to train face models vary little or unevenly in pose. In order to solve the pose-invariant face recognition problem, proposes a method based on metric learning, CPP Loss. This method can effectively utilize the limited pose varieties in the training set and further improve the performance of benchmark model under cross-pose conditions.