AN EFFICIENT APPROACH TOWARDS IDENTIFYING AND RECOGNIZING FACES IN REAL TIME EMPLOYING DEEP LEARNING

Abstract

Face recognition can be termed as one of the most common uses of computer vision and image processing, in which a computerised system automatically recognises a person's face from a big image collection or even a live video. This thesis focuses on facial recognition, a topic that has received a lot of attention due to its usefulness in a variety of civilian and military applications. These systems are being used for numerous objectives like fraud management, security etc. and improving user experience. Face recognition algorithms have been developed in a variety of ways, with varying degrees of success. Due to the dynamic nature of the human face and the various stances it might adopt, this challenge is difficult to solve. In the present project, we propose to use YOLO which utilizes fewer samples as compared to CNN as the initial object detection method. We have used transfer learning on YOLO to use it for detecting faces. For finetuning face recognition models, we suggest a transfer learning (TL) method which combines TL techniques with CNN. Transfer learning may be used to train long-lasting, top-performance ML models that need less time and resources as compared to models learnt from the ground up. We executed transfer learning on a pretrained face recognition model to create a network capable of generating correct predictions on considerably smaller datasets. To achieve acceptable accuracy, convolutional neural networks are known to require big datasets. This project presents a solution to this problem by minimising the number of people involved, thereby reducing the number of training examples to one while maintaining near perfect accuracy applying the transfer learning principle