Pose and Illumination Invariance Using Efficient Discriminant Component Analysis with Multi Class Support Vector Machine In Face Recognition System

Abstract

ABSTRACT newlineFace is a complex multidimensional structure and needs a good computing technique for recognition. While using automatic system for face recognition, computers are easily confused by changes in illumination, variation in poses and changes in angle of faces. A Numerous techniques are being used for security and authentication purposes which includes areas in detective agencies and military purpose. The basic idea behind the automatic face recognition system under the condition of pose and illumination is to decrease the Equal Error Rate (EER) and increase the Recognition Rate (RR). By using the feature extraction methods and dimensionality reduction techniques in the pattern recognition applications, facial recognition systems has been produced with distinct measure of success. This can be achieved by overcoming the issues such as dimensionality reduction while storing the features subspace of training and testing set and by using effective classifier at the recognition phase. newlineFeature extraction plays an major role to provide an efficient way of feature subspace extraction with which benefited from wavelet decomposition and Eigen faces method and is based on Principal Component Analysis (PCA)[64][65] and Linear Discriminant Analysis with proposed decision rules. After generating feature vectors, distance classifier and multi-class Support Vector Machines (SVMs) are used for classification step. To examine the classification accuracy according to different dimension of training set, chosen feature extractor by decision rules and chosen kernel function for SVM classifier are classified. newlineThe proposal presents a novel 2D image-based approach that can simultaneously handle illumination and pose variations with Efficient Discriminant Component Analysis (EDCA) to enhance face recognition rate. It is much simpler, requires much less computational effort than the methods based on 3D models, and provides a comparable or better recognition rate. The illumination variation problem can be over come by pre processing