Dimensionality Reduction using Eigenfaces and Fisherfaces for Face Recognition Applications using Kernel LMS Algorithm

Abstract

Abstract
Face recognition is an intricate numerical technology in computer systems that can support recognition of human faces using methods such as principal component analysis (PCA), linear discriminant analysis (LDA), etc. by comparing the given facial uniqueness with the already available face database. PCA and LDA are classical feature extraction and data representation techniques used in face recognition algorithm involving unsupervised statistical methods. The 2-D facial image can be transformed into a 1-D vector of pixels and projected into the principal parts of the feature space called the eigen space projection that
best encodes the variations among the known faces. The projections of the given face are then compared with the available training set and the face is identified. A novel approach of dimensionality reduction of the covariance matrix and weight update using the kernel LMS algorithm has also been dealt. Euclidean distance, recognition rate, eigenvalue variation, eigen space scatter and computational time are considered as dependable measures to evaluate the face recognition algorithms based on eigenfaces and fisherfaces. Three methods
were chosen and implemented to test their efficiency over standard Olvetti research laboratory (ORL), YALE databases containing large number of frontal face images of different face classes. One of the earliest and most popular methods, the eigenfaces method, which concentrates on maximizing the total scatter across the face subspace was observed to have performed less satisfactorily out of the three methods tested. The Fiske, faces method has
much better performance in which the within-class scatter is minimized, while increasing the between-class scatter and the generalized discriminant analysis (GDA) method, which tries to find the null space of the within-class scatter.

Keywords: Dimensionality reduction, eigenfaces, face recognition, facespace, fisherfaces, generalized discriminant analysis, linear discriminant analysis, principal component analysis

Cite this Article
Raju Dara, Ch. Satyanarayana, Govardhan A. Dimensionality Reduction using Eigenfaces and Fisherfaces for Face Recognition Applications using Kernel LMS Algorithm. Journal of Computer Technology & Application. 2015; 6(2): 29–55p.