Performance Analysis of Local Binary Pattern Features with PCA for Face Recognition

Abstract

Objective: To propose the robust face recognition system in front of two different challenges, pose variations and illumination variations. Methods/Statistical Analysis: Five different methods are used for feature extraction (1) Local binary pattern histogram (LBPH) from entire image (2) LBPH from image divided into 9 different regions (3) Local binary pattern image (LBPI) (4) Gabor features (GF) and (5) Two dimensional discrete wavelet transform (2D-DWT) using haar-3 wavelet. For the last four methods principal component analysis (PCA) is used for dimensionality reduction and classification is performed by two non linear functions, radial basis function (RBF) and polynomial function (PF) based on support vector machines (SVM's). Findings: Application/Improvements: For performing the experiments two databases are used, ORL face database and extended yale-B face database. For the former database it is the LBPH feature extracted from different facial regions outperforms the other features and for the second database LBPI feature outperforms the other features.