基于深度学习MPCANet的年龄估计

Age Estimation Based on Deep Learning MPCANet

提出了一种基于多层PCA网络(MPCANet)的深度学习模型来进行年龄估计.它是基于卷积神经网的结构来设计的,并且用来提取年龄特征.MPCANet是主成分分析网络(PCANet)的一种改进,它是最近提出的一种深度学习算法,MPCANet模型结构组成的成分:(1)卷积滤波层是采用多层级联主成分分析(PCA),(2)非线性层则采用二进制哈希,(3)特征抽取层使用直方图统计方法.使用核支持向量回归(K-SVR)进行估计年龄值.实验分别在两个数据库(FG-NET and MORPH)上进行,实验结果表明该方法比目前最新的方法表现得更好.

This paper investigates deep learning techniques for age estimation based on the multi principal component a- nalysis network（ MPCANet）. A new framework for age feature extraction based on deep learning model with eonvolutional neural network（CNN） is built. The MPCANet is a variation of principal component analysis network（PCANet） ,which is recently proposes deep learning algorithms. The MPCANet model architecture components ： （ 1 ） the use of Multi cascaded principal component analysis （PCA）in the convolution filter layer;（2） the nonlinear process layer by binary hashing;and （3） the use of block histogram in the feature pooling layer. We use K-SVR（ Kernel function Support Vector Regression, K-SVR） for age estimation. Experimental results on two datasets（ FG-NET and MORPH）show that the proposed approach is significantly better than the state-of-the-art.