基于对比学习的多特征融合戴口罩人脸识别

随着计算机视觉技术应用的发展和智能终端的普及，口罩遮挡人脸识别已成为人物身份信息识别的重要部分。口罩的大面积遮挡对人脸特征的学习带来极大挑战。针对戴口罩人脸特征学习困难这一问题，提出了一种基于对比学习的多特征融合口罩遮挡人脸识别算法，该算法改进了传统的基于三元组关系的人脸特征向量学习损失函数，提出了基于多实例关系的损失函数，充分挖掘戴口罩人脸和完整人脸多个正负样本之间的同模态内和跨模态间的关联关系，学习人脸中具有高区分度的能力的特征，同时结合人脸的眉眼等局部特征和轮廓等全局特征，学习口罩遮挡人脸的有效特征向量表示。在真实的戴口罩人脸数据集和生成的戴口罩人脸数据上与基准算法进行了比较，实验结果表明所提算法相比传统的基于三元组损失函数和特征融合算法具有更高的识别准确率。

Multi feature fusion for masked face recognition based on contrastive learning

With the development of computer vision technology and the popularization of intelligent terminals, facial recognition under mask occlusion has become an important part of character identity information recognition. The large area occlusion of masks poses great challenges to the learning of facial features. To solve this problem, this paper proposed a multi feature fusion based masked face recognition algorithm based on contrastive learning. This algorithm improved the traditional face feature vector learning loss function based on the triple relationship. It proposed a loss function based on the multi-instance relationship, which fully excavated the intra-modal and inter- modal correlation between multiple positive and negative samples of the masked face and the full face. Then, the features with high discrimination ability were learnt from the face. Meanwhile, it combined the local features such as eyebrows and eyes, as well as global features such as contours, to learn the effective feature vector representation of the masked face. This paper compared it with the benchmark algorithm on real masked face datasets and generated masked face data. The experimental results show that the proposed algorithm has higher recognition accuracy than the traditional triple loss function and feature fusion model.