Face Recognition using a Newly Developed Linear Subspace Learning Method

Abstract

Face recognition is considered a specific physiological biometric in order to identify an individual according to the physical features of the human face. Much research has been conducted in such areas, but still more accurate processes are required for biometric facial recognition. This article presents a novel linear subspace learning method for face recognition which not only can take advantage of principle components analysis (PCA) as a successful feature extraction, but also can apply nearest local centroid mean vector (LMKNCN) as an effective classifier to improve the classification performance. The main goal of this particular scheme is to handle two common existing issues in recognition techniques: named sensitivity to the training sample size and negative effects of outliers. Moreover, to illustrate the performance of proposed developed PCA, we compare it with the latest dimensionality reduction techniques such as traditional PCA and KPCA on publicly available face dataset. Experimental results illustrate that our newly developed method has significantly achieved better performance over the same face database compared with the former KNN-based algorithms.