关键词: 关键词: 非受限人脸年龄分类, 深度残差网络, 迁移学习, ImageNet, IMDB-WIKI-8, Adience

Abstract: Abstract: Concern the problem that the accuracy of age classification of face images under unrestricted conditions is low, a new method of face age classification under unconstrained conditions based on deep residual network and large datasets pre-training was proposed. Firstly, the deep residual network was used as the basis convolutional neural network model to deal with the problem of face age classification. Secondly, the deep residual network was trained on the ImageNet dataset to learn the expression of basic image features. Thirdly, The IMDB-WIKI with large-scale face age images was cleaned, and the IMDB-WIKI-8 dataset was established for fine-tuning the deep residual network, and achieved the general objects to the face age images migration learning, thus making the model adapt to the distribution of the age group and improve the network learning capability. Finally, the fine-tuned network model was used to train and test on the unconstrained Adience dataset, and the age classification accuracy rate was obtained with the cross-validation method. Through the comparison of 34/50/101/152 residual networks, it can be seen that the higher the accuracy of the classification of the deeper layers of the network, and a new state-of-the-art age classification result on Adience dataset with the accuracy 65.01% was achieved using the 152 residual network. The experimental results show that the combination of deeper residual network and large data set can effectively improve the accuracy of face age classification.

Key words: Keywords: unconstrained face age classification, deep residual network, migrate learning, ImageNet, IMDB-WIKI-8, Adience