Feature extraction using PCA and Kernel-PCA for face recognition

The face recognition system consists of a feature extraction step and a classification step. In this paper, the researcher studies the use of linear and nonlinear methods for feature extraction in the face recognition system. The linear Principal component analysis (PCA) which is widely used in the face recognition is used to construct the feature space and extract features. The Kernel-PCA is extended from PCA to represent nonlinear mappings in a higher-dimensional feature space. Several parameters of Kernel functions are investigated and expected to affect the recognition performance. The k-nearest neighbor classifier with Euclidean distance is used in the classification step. Our experiments are carried out on the ORL face database which contains variability in expression, pose, and facial details. Experimental results show that Kernel-PCA with Gaussian function can give a correct recognition rate similar to PCA and higher than Kernel-PCA with polynomial function.