Comparison of PCA and Nonlinear PCA for Face Recognition

W. Huang and H. Yin (UK)


Dimensionality reduction, linear and nonlinear, facerecognition, principle component analysis


Dimensionality reduction greatly facilitates pattern classification. Various techniques, linear and nonlinear, have been widely proposed and used for dimensionality reduction in face recognition systems. Principle Component Analysis (PCA) has proved to be a simple and efficient linear method; while many nonlinear methods such as kernel PCA, have been proposed recently. The purpose of this paper is to evaluate the performance of the linear PCA and several popular nonlinear PCA methods for dimensionality reduction in face image classification. The experiments are conducted on a real-world face database. The results show that both PCA and nonlinear methods yield good performances. The differences in recognition rates between them however are smaller than 1% in all implementations. In other words, the performance gain by the computational demanding nonlinear PCA methods is insignificant. Thus this makes linear PCA a better choice than nonlinear methods in face recognition for its simplicity.

Important Links:

Go Back