Probabilistic Two-dimensional Principal Component Analysis
概率二维主分量分析

Two-dimensional principal component analysis(2DPCA)is an approach to feature extraction and dimen- sionality reduction for an image represented straightforward as a matrix.In this paper,a probabilistic model for 2DPCA, called P2DPCA,is proposed.First,the principal components(vectors)are derived through maximum-likelihood estima- tion of parameters in the generative probabilistic model.Then,due to dealing properly with missing data,we present an expectation-maximization(EM)algorithm for estimating the parameters of the model and principal components.The application to cluster face images using mixtures of P2DPCA models shows that P2DPCA model can be a tool for density-estimation of image matrix.Experimental results on face image reconstruction with missing data illustrate the effectiveness of the model and the EM iterative algorithm.