结合主元成分分析的受限玻耳兹曼机神经网络的降维方法

Dimensionality Reduction Method Based on Restrained Boltzmann Machine Neural Network with Principal Component Analysis

介绍一种能够有效地获取数据本质的基于受限玻耳兹曼机(RBM)神经网络的降维(RBMNNBDR)方法,并结合主元成分分析法(PCA),提出了一种新颖的复合特征降维方法,即PCA-RBMNNBDR.结合人脸研究的几个应用示例,通过实验对PCA-RBMNNBDR、RBMNNBDR和线性判别式分析(LDA)方法进行比较.结果表明,PCA-RBMNNBDR方法在人脸图像降维和分类方面有更好的效果,其分类正确率高于RBMNNBDR和LDA方法.

A novel restrained Bohzmann machine neural network based dimensionality reduction （RBMNNBDR） method was introduced. It is improved by combining it with principal component analysis （PCA） to generate a hybrid dimensionality reduction method called the PCA- RBMNNBDR. Then, it is compared with the RBMNNBDR method and LDA method by applying them in face research. The experiments show the PCA-RBMNNBDR method outperforms both RBMNNBDR method and LDA method; furthermore it can efficiently capture the intrinsic manifold structure of the face images.