一种融合Resnet-50与SVR的年龄估计

An Age Estimation Combining Resnet-50 and SVR

为提高在非受限条件下的人脸图像年龄估计的精度,提出一种融合残差网络与支持向量回归的年龄估计方法(Resnet-50_SVR)。Resnet-50网络是一种解决神经网络结构退化问题的成熟算法,使用Resnet-50网络提取面部更佳的年龄特征,即使用Resnet-50结构在ImageNet数据集上对网络模型进行预训练,再将预训练的模型在MORPH数据集上进行微调,并结合SVR算法在FG-NET数据集上进行测试。实验结果表明,Resnet-50_SVR对人脸图像的年龄估计效果良好。

In order to improve the accuracy of age estimation of face images under unrestricted conditions, proposing an age estimation method that combines residual network and support vector regression(Resnet-50_SVR). Resnet-50 network is a mature algorithm to solve the problem of degenerate neural network structure, using the Resnet-50 network to extract better age features of the face, that is to say, the network model was pretrained by Resnet-50 structure on the ImageNet dataset, and finetuned on the MORPH dataset, and combined with the SVR algorithm on the FG-NET dataset for testing. The experimental results demonstrated that Resnet-50_SVR is effective in age estimation of face images.