Implementation of Convolutional Neural Network with Optimization Techniques for Face Recognition

Face images is the key factor though which an individuals are distinguished. Because of current pandemic impact of COVID-19 the entire universe is drawing closer towards contact less Gadgets where less physical collaboration is required. In view of this, we have proposed the Convolutional Neural Network (CNN) model and applied on various face databases by measuring the behaviour of various optimizers. The experimentation has been performed on various face databases. In particular, AR, LFW, LAG, DFW, UKFace Database, PSD, ASPS. These databases contains the facial images with variations in illumination, expression, various poses of face image, occluded images by face mask, sunglasses, beared, mustache also contains face images of different age group. Furthermore, the proposed CNN model have evaluated on pre and post plastic surgical face images. The working of some of the optimizers listed as, PowerSign, AddSign, RMSprop, Adam are observed on these face images. From the said optimizers, PowerSign and AddSign perform better for sequential data. However, it comes to 2D pictures its performance diminishes to incredible expand. When performed an experimentation with RMSprop achieved better accuracy. Nevertheless, suffers from the local minimum. On a contrary, Adam outperform with respect to CNN model by obtaining local minima in less time. Adam achieves 98% to 90% of accuracy for AR to PSD(Plastic surgery face database) and ASPS(American Society of Plastic Surgeon) database.