基于注意力机制和深度恒等映射的人脸识别

Face recognition based on attention mechanism and deep equivariant mapping

针对在自然场景下,受到姿态变化、侧脸等因素的干扰,人脸识别难度提升的问题,提出一种改进的残差神经网络,优化原有网络模型结构并且引入加强通道重要特征获取的注意力机制。同时添加深度残差恒等映射模块,不增加过多参数量的情况下,实现了在深度特征空间层将侧脸特征映射为正脸特征的功能,进一步实现人脸识别率。实验结果表明:在常用的LFW,CFP,IJB-A等人脸数据集上,提出的网络优于现有的残差神经网络,提升了非约束条件下的人脸识别效果。

Aiming at the problem that in natural scenes,face recognition becomes more difficult due to the interference of posture changes,side faces and other factors,an improved residual neural network is proposed,which optimizes the original network model structure and introduces an attention mechanism to enhance channel important feature acquisition.At the same time,depth residual equivariant mapping module is added,and the function of mapping the side face features into the positive face features in the depth feature space layer is realized without adding too many parameters,further,realize human face recognition rate.The experimental results show that the proposed novel neural network is better than the existing residual neural network on commonly used face datasets such as LFW,CFP,IJB-A,etc.,and improves the face recognition effect under unconstrained conditions.