Biometric Authentication using lightweight Convolutional Neural Network

Abstract

Biometric technology like face recognition, retina scan and finger print mapping are very useful in identification and verification to recognize a person. In recent years, many deep learning algorithms have proposed for various applications specially face recognition. Hierarchical architecture of deep learning methods significantly improves the Face Recognition performance and encourages diverse and efficient real – world application. In Face Recognition Technology mainly four phases are exist: face detection, face alignment, feature extraction and classification. Many challenges are faced in face recognition like: pose, age, illumination, gender etc. Convolutional neural network (CNN) achieves good performance in face recognition system. In Deep learning especially convolutional neural network models required large data sets to train the deep layers and provide more accuracy. In CNN, Many models work for pattern recognition. One of them is Visual Geometry Group (VGG), which have large number of layers in architecture. In this research, light VGG – 16 models is proposed to recognize faces. The proposed model achieves good accuracy and low loss value in comparison to traditionally VGG -16 and comparatively reduces the number of parameters because large number of parameter takes significant memory size.