融合注意力机制的人脸识别算法研究

Research on face recognition algorithm integrating attention mechanism

FaceNet人脸识别算法是目前较为主流的人脸识别算法,其运行速度快被广泛应用于工业界。针对FaceNet人脸识别网络存在对面部遮挡人脸识别精度低的问题,提出了一种融合注意力机制的FaceNet人脸识别算法。该算法在FaceNet的基础上引入GhostNet特征提取网络对人脸更好的提取人脸特征,并融合注意力机制与特征金字塔(feature pyramid networks,FPN)加强特征提取网络实现对3种尺度特征图中局部信息的放大,加强不同感受野下的特征提取,增强较为重要的特征信息。实验结果表明,提出的人脸识别算法取得了良好的识别效果,在人脸数据集(LWF)下准确率达到99.62%。对有遮挡的面部识别也取得了较好的检测结果,可准确识别存在遮挡的人脸目标。

FaceNet face recognition algorithm is currently the mainstream face recognition algorithm, its fast running speed is widely used in the industry. FaceNet face recognition network has the problem of low accuracy in face occlusion face recognition. This paper presents a FaceNet face recognition algorithm that combines attention mechanism. Based on FaceNet, GhostNet feature extraction network is introduced to extract face features better. Feature pyramid networks(FPN) is combined to enhance feature extraction network to enlarge local information in three scale feature maps, enhance feature extraction under different perception fields and enhance more important feature information. The experimental results show that the face recognition algorithm in this paper achieves good recognition results, and the accuracy reaches 99.62% in the face dataset(LWF).Better detection results are also obtained for face recognition with occlusion, which can accurately identify face targets with occlusion.