融合人脸五官信息的深度年龄估计

摘要

The paper presents a new mode of solution for deep age estimation by facial features auxiliary,which fuses the traditional facial information with the convolutional neural network(CNN)to achieve the age estimation,in order to reinforce the generalization ability of system model. The solution estimates age from image pixels directly,which makes the locally aligned face image block generated by the key points of the face as the input of the CNN.The system improves the performance significantly by using the multi-scale CNN network structure. At the same time,it apply the traditional method to strengthen the information of facial areas. The experiments on MORPH AlbumⅡillustrate the superiorities of the proposed method over other state-of-the-art methods.%本文提出了一种新型的基于人脸五官辅助的深度年龄估计方法,将传统的人脸五官区域特征提取加分类器设计方法与基于深层卷积神经网络(convolutional neural network,CNN)的端到端分类方法进行融合来解决年龄估计问题,增强了系统模型的泛化能力.该方法将面部关键点生成的局部对齐的人脸图像块作为CNN的输入,直接从图像的像素点评估年龄,采用多尺度分析网络结构极大地提高了性能,同时又利用传统算法增强了五官区域的信息.最后通过在MORPH AlbumⅡ上的实验表明文中提出方法比其他同类研究方法更加优秀.