An Individual Local Mean-based 2DPCA for Face Recognition under Illumination Effects

Abstract

Principal component analysis (PCA) is a classical technique in pattern recognition and computer vision. It is one of the most successful techniques for face recognition. The PCA consists of two main steps including (I) covariance matrix calculation and (II) eigenvector and eigenvalue extraction. In case of face recognition, the input image is converted to the vector form before forwarding to covariance matrix computation. Then, the matrix is used to extract the eigenvector and eigenvalue. Two-dimensional PCA (2DPCA) is introduced to reduce high-dimensional problems. The illumination effect problems in the face recognition is still needed to be resolved. In order to improve and solve the problems, this paper proposes an individual local mean-based 2DPCA (ILM-2DPCA), which replaces a single local mean in 2DPCA method. The individual local mean can provide more appropriate mean to each image, which can reduce the illumination effect effectively. The experimental study is set up on dataset Yale face database B+. The results indicate that the proposed method outperforms, based on the accuracy rate, all the baseline methods which are 2DPCA, I-2DPCA, Bi2DPCA and 2D²PCA.