摘要
由于在特殊场景下大量标注人脸数据样本识别时需要大量带有身份标记的训练样本,且无法精准提取小样本特征,故提出单样本学习(one-shot learning)的人脸识别算法。选取并赋值单样本人脸图像像素点中间值,保存至缓冲区进行遍历,利用Siamese Network模型计算遍历结果共享权重,利用共享权值识别图像特征相似性,得到人脸识别结果。结果表明,与基于卷积神经网络的人脸识别方法相比,所研究方法识别准确率达到95.68%,识别效率达到354.25 s,结果更好。由此说明所研究方法在小样本的情况下也能更为快速且准确地完成人脸识别任务。
Because a large number of training samples with identity marks are needed when labeling a large number of face data samples for recognition in a special scene,and the features of small samples can not be accurately extracted,a face recognition algorithm based on one⁃shot learning is proposed.The intermediate value of the pixel points in facial image of a single sample are selected and assigned.And then,the value is saved at the buffer for traversal.The traversal results and the shared weight are calculated by Siamese Network model.The shared weight is used to identify the image feature similarity to obtain the face recognition results.The results show that the recognition accuracy rate of the studied method is 95.68%and its recognition efficiency is 354.25 s,which is better than that of the face recognition method based on convolutional neural network(CNN).In addition,this method can accomplish the task of face recognition more quickly and accurately in the case of small samples.
作者
程远航
余军
CHENG Yuanhang;YU Jun(College of Science and Technology,Guizhou University,Guiyang 550003,China)
出处
《现代电子技术》
2021年第19期76-80,共5页
Modern Electronics Technique
基金
贵州省科技合作计划项目(黔科合LH字[2017]7227号)
贵州大学2017年度学术新苗培养及创新探索专项(黔科合平台人才[2017]5788号)。
