基于深度学习的人脸识别算法研究
首发时间:2017-12-28
摘要:传统的人脸识别算法主要是基于图像的浅层特征提取,比如LBP、SIFT、HOG等图像特征描述算子,然后进行多种浅层特征融合,PCA降维之后再采用传统的机器学习分类器进行人脸识别,比如线形判别式、SVM、联合贝叶斯分布等。深度学习能够提取图像的高层的、抽象的、概念化的特征,能够从图像全局的角度提取到人脸最本质的特征,全局的特征能够很好的解决人脸的姿态、表情、光线、年龄、遮盖等噪声影响。本文的的创新性体现在使用时下高效Inception-ResNet和SqueezeNet网络模型和多种损失函数的结合,设计出高效的人脸高层特征提取器,并且直接在LFW公开测试集上测试,取得了很高的性能。?
关键词: 人工智能 人脸识别 卷积神经网络 损失函数?????
For information in English, please click here
RESEARCH ON FACE RECOGNITION ALGORITHM BASED ON DEEP LEARNING
Abstract:The traditional face recognition algorithm is mainly based on the image of the shallow feature extraction, such as LBP, SIFT, HOG image feature descriptors, and then a variety of shallow features fusion, PCA dimension reduction and then use the traditional machine learning classifier for face recognition, such as linear discriminant, SVM, Joint Bayesian distribution. Deep learning can extract the high-level, abstract and conceptual features of the image, and can extract the most essential features of the human face from the perspective of the image. The global features can well solve the problems of facial pose, expression, light, age, cover and other noise effects. The innovation of this paper is to design a highly efficient facial feature extractor using the combination of the current and efficient Inception-ResNet and SqueezeNet network models with a variety of loss functions and to test directly on the LFW public test set, achieving very high Performance.?????
Keywords: artificial intelligence;face recognition; CNN; loss function
基金:
引用
No.****
动态公开评议
共计0人参与
勘误表
基于深度学习的人脸识别算法研究
评论
全部评论0/1000