作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程 ›› 2018, Vol. 44 ›› Issue (5): 246-251. doi: 10.19678/j.issn.1000-3428.0046059

• 图形图像处理 • 上一篇    下一篇

基于深度学习与随机森林的人脸年龄与性别分类研究

董兰芳,张军挺   

  1. 中国科学技术大学 计算机科学与技术学院,合肥 230027
  • 收稿日期:2017-02-22 出版日期:2018-05-15 发布日期:2018-05-15
  • 作者简介:董兰芳(1970—),女,副教授,主研方向为图像处理、计算可视化、视频图像智能分析;张军挺,硕士研究生。

Research on Face Age and Gender Classification Based on Deep Learning and Random Forest

DONG Lanfang,ZHANG Junting   

  1. College of Computer Science and Technology,University of Science and Technology of China,Hefei 230027,China
  • Received:2017-02-22 Online:2018-05-15 Published:2018-05-15

摘要: 为提高在非限制性环境下的人脸年龄估计与性别识别准确率,提出一种基于深度卷积神经网络的人脸特征提取方法。通过采用一般到特殊的微调方案,在大规模数据集上进行人脸识别预训练得到的VGG-Face模型,运用该模型在CelebA人脸属性数据集上对其中5个属性进行微调训练,得到人脸属性模型,将网络全连接层特征进行连接作为人脸特征向量。使用随机森林分类器在Adience数据集上进行训练和测试,利用随机森林方法处理高维的数据,选出对年龄与性别分类较重要的特征。实验结果表明,该方法能够克服复杂光照、姿态变化的影响,准确地对自然场景下的人脸进行年龄估计和性别识别。

关键词: 年龄估计, 性别识别, 深度卷积神经网络, 微调训练, 随机森林

Abstract: In order to improve the accuracy of face recognition and gender recognition in a non-restrictive environment,a facial feature extraction method based on deep convolution neural network is proposed.The VGG-Face model,which is pre-trained by face recognition on a large-scale dataset,is used to fine-tune five of the five attributes on the CelebA face attribute dataset by using a general to special fine-tuning scheme.Face attribute model is gained,and it connects the network full connection layer features as the facial feature vector.Training and testing are conducted on the Adience dataset using a random forest classifier,and high-dimensional data are processed using a random forest approach to select features that are more important for age and gender classification.Experimental results show that this method can overcome the influence of complex illumination and pose changes,and can accurately estimate the age and gender of human face under natural scenes and achieve good effects.

Key words: age estimation, gender recognition, deep convolutional neural network, fine-tuning training, random forest

中图分类号: