Intelligent Biometric System using PCA and R-LDA
Anupam Shukla, Joydip Dhar, Chandra Prakash, Dhirender Sharma, Rishi Kumar Anand, Sourabh Sharma · 2009
The paper presents a novel biometric authentication approach using principal component analysis (PCA), regularized-linear discriminant analysis (R-LDA) and supervised neural networks. Low dimensional feature vectors of human face images are required to drive neural networks effectively. After histogram equalization process each image is presented to PCA or R-LDA for normalization and dimension reduction. The preprocessing steps of PCA or R-LDA produce Low dimensional feature vectors appropriate for training. Neural network has a great deal of nerve cell and can accomplish parallel distributing operation. Backpropagation (BP), radial basis function(RBF) & learning vector quantization (LVQ) are used as classifiers. The analysis of obtained results shown that R-LDA preprocessed feature vectors driven by supervised neural networks are having better recognition performance than PCA. While among supervised neural networks RBF gave most matched output during testing.