基于卷积神经网络的人脸年龄估计方法

Facial Age Estimation Method Based on Convolutional Neural Network

针对年龄估计算法中使用单一标签编码方式的问题,提出一种基于高斯分布的标签分布取代原有的编码方式,通过这种方式,一张人脸图像不仅可以用于它本身年龄的学习,也可用于其相邻年龄的学习。设计一种与之相适应的卷积神经网络,通过标签的概率分布与输出的类别概率之间的K-L距离对网络进行训练,网络结构较小,所需训练时间更短,即使在训练图片数量有限的情况下依然能保证较好的准确率。

Aiming at the problem of single-label coding method in facial age estimation,label distribution based on Gaussian distribution is proposed to replace the original coding method. In this way,a face image can be used not only for its own age learning,but also for its adjacent age learning. A convolutional neural network is proposed to train the network by the Kullback-Leibler divergence distance between the label distribution and the probability of the output. The network is small and needs shorter training time,even in the case of limited number of training images can still guarantee a better accuracy rate.