基于双通道FaceNet的遮挡人脸识别OA
Occluded face recognition based on dual-channel FaceNet
在大面积人脸遮挡情况下,传统 FaceNet 可利用的图像特征信息大幅度减少.为解决这一问题,该文研究了基于双通道 FaceNet 的遮挡人脸识别方法.该文创新性地构建了双通道FaceNet 网络模型,该模型包含全局通道与局部特征通道.该方法加强了网络对局部特征的学习,提高可用特征的利用率,从而改善了 FaceNet 在有遮挡情况下的识别准确率.在没有遮挡的情况下,双通道 FaceNet 比传统的 FaceNet 识别准确率提高了 0.47%.在有遮挡的情况下,双通道 FaceNet 的识别准确率提高了 5.10%.在模拟人脸相互遮挡的场景中,当遮挡比例为 40%和60%时,双通道 FaceNet 的识别准确率分别提高了 3.99%和 9.09%.实验结果证明了双通道FaceNet 对遮挡人脸识别的有效性.
In the case of large area face occlusion,the available image feature information of traditional FaceNet is greatly reduced.To address this problem,occluded face recognition method based on dual-channel FaceNet is proposed.This paper innovatively constructs a dual-channel FaceNet network model,which includes global channel and local feature channel.This method can enhance the network's learning of local features,improve the utilization of available features,and improve the recognition accuracy of FaceNet in the case of occlusion.In the case of no occlusion,the dual-channel FaceNet improves the recognition accuracy by 0.47%compared with the traditional FaceNet,while in the case of occlusion,the recognition accuracy is improved by 5.10%.In the scenario of simulating mutual occlusion of faces,when the occlusion ratio is 40%and 60%,the recognition accuracy of dual-channel FaceNet is improved by 3.99%and 9.09%,respectively.Experimental results demonstrate that the effectiveness of dual-channel FaceNet for occluded face recognition.
任星光;丁晨寅;段纳;孙浩
江苏师范大学 电气工程及自动化学院,江苏 徐州 221116江苏师范大学 电气工程及自动化学院,江苏 徐州 221116江苏师范大学 电气工程及自动化学院,江苏 徐州 221116江苏师范大学 电气工程及自动化学院,江苏 徐州 221116
信息技术与安全科学
遮挡人脸FaceNet网络双通道局部特征欧氏距离
occluded faceFaceNet networkdual-channellocal featureEuclidean distance
《南京理工大学学报（自然科学版）》 2026 (2)
152-160,9
国家自然科学基金(62173166)
评论