Sparse Kernel Principal Components Analysis for Face Recognition in RGB Spaces

Abstract

This paper presents a kinds of information fusion algorithm based on multi-channel color image. The color face image is first separated into three pseudo grayscale images: R, G, and B, then the partial characteristics of face is extracted by use of Gabor wavelet transform from each component to be eigenvector in series connection, which will be through dimensionality reduction by sparse kernel principal components to be recognized and classified by the nearest classifier. In order to testify the validity, we make experiment by use of XM2VTS color face dataset and the experimental result supports the proposed method.