核主成分分析网络的人脸识别方法

Kernel principal component analysis network method for face recognition

主成分分析网络(principal component analysis network,PCANet)是当前流行深度学习模型,在人脸识别、目标识别、纹理分类和手写体数字识别等方面有广泛应用。在PCANet方法基础上,该文提出基于核主成分分析网络(kernel principal component analysis network,KPCANet)的人脸识别方法。算法由四部分组成:主成分分析(PCA)、核主成分分析(KPCA)、二值化哈希、分块直方图。在Extended Yale B和AR两个经典人脸库上的实验结果表明,所提方法在识别性能上优于PCANet方法 ,算法对于不同光照、表情变化下的人脸有更好的识别率和鲁棒性。

Principal component analysis network （PCANet） is a popular deep learning classification method, which has caused wide attention in the area of computer vision due to its practical applications in face recognition, hand-written digit recognition, texture classification, and object recognitions. On the basis of PCANet. The kernel principal component analysis network （KPCANet） method is proposed for face recognition. The model is constructed by four processing components, including principal component analysis （PCA）, kernel principal component analysis （KPCA）, binary hashing, and block-wise histograms. The performance of the proposed method is evaluated using two public face datasets, i. e. , Extended Yale B database and AR face database. The results show that KPCANet outperforms PCANet method. Especially when the face images have large variations about illuminations and expressions, KPCANet gives better recognition results.