Feature Extraction Based on Mixture Probabilistic Kernel Principal Component Analysis

Feature extraction of training samples and testing samples face the problem of the high non-linear by complexity of the distribution of the samples. In contrast to linear PCA, KPCA is capable of capturing part of the higher-order statistics which are particularly important for encoding image structure. The probabilistic kernel principal component analysis (PKPCA), defines PPCA probability model by non-linear mapping in the high-dimensional feature space. This paper presents the mix model of the probability of kernel principal component analysis (MPKPCA) method, which adopt a non-linear mapping to make the data from low-dimensional space to the high-dimensional kernel space, in kernel space, using the mixed probability principal component analysis (MPPCA), it combines the advantages of kernel principal component analysis (KPCA) and MPPCA characteristics. Experimental results under complex scenery demonstrate that the proposed algorithm is feasibility and effectiveness.