基于深度卷积网络的人脸年龄分析算法与实现

摘要

作为一种新兴的生物特征识别技术，基于人脸图像的年龄估计技术在目前已经成为计算机视觉、人机交互等领域的一个重要研究课题。2006年以来，深度卷积网络在图像识别、语音识别和自然语言处理等领域广泛使用，取得了很好的效果。本文基于深度卷积网络的人脸年龄分析算法，构建一个多层卷积神经网络，通过卷积神经网络获取深度卷积激活特征，作为人脸年龄估计的特征，并利用支持向量机(SVM)的方法训练年龄估计模型，得到年龄估计结果，在人脸识别权威数据集Morph上获得了91.3%的正确率，同时也对比在了不同条件下对实验结果的影响。%As an emerging biometric technology,the age estimation based on facial images has become an important research topic in the field of computer vision,human computer interaction,etc.Since 2006,Deep Convolutional Neural Network has been widely used and obtained good effects in the field of image recognition,speech recognition,natural language processing,etc..In age estimation algorithm based on Deep Convolutional Neural Network,the paper constructs a multilayered convolutional neural network.Taking the deep convolutional activation features as the facial distinctions of age estimation, the paper applies Support Vector Machine(SVM) to train the age estimation model and acquire estimation results. The algorithm has obtained 91.3% accuracy in Morph,the authoritative facial recognition dataset.Meanwhile,the experiment results in different conditions have been compared.