Main Article Content
Abstract
Face Recognition is an important basic primitive for various applications and fields such as threat detection, immigration, educational institutions, video analysis & surveillance, sport and biometrics etc. In face recognition, face detection is first step to find out the region of face and this step suffers from variations while determining bounding box for face. This leads to the misalignment in face regions across the samples of intra-class and inter-class faces. In this paper, we have proposed to use convolutional neural network (CNN) to overcome this issue. CNN is a one of popular technique of deep learning. We compared proposed approach of generating training samples to be included in training of CNN such that it improves the recognition accuracy significantly as compared to the one when only original training samples are used in training.