A Proposed novel approach to face recognition using CNN

Abstract

Facial recognition has emerged as a crucial technology with applications spanning from security and surveillance to user authentication and human-computer interaction. This work is a comprehensive study of facial recognition techniques leveraging Convolutional Neural Networks (CNN), which is a class of deep learning models known for their exceptional performance in image processing tasks. The primary objective of my research is to develop a strong facial recognition system capable of accurately identifying individuals across various real-world scenarios and challenges. The thesis begins by providing an overview of the fundamentals of CNN, their architecture, and their relevance in image-based pattern recognition. Then it delves into the pre-processing steps involved in preparing facial images for CNN-based recognition, including data collection, data augmentation, and face detection. Special attention is given to handling occlusion, illumination variations, and pose changes often encountered in real-world environments. The core of my work focuses on the design and implementation of CNN-based facial recognition models. Different CNN architectures are explored, and their performance is evaluated using benchmark datasets. In this research 21000 images from Kaggle as the dataset are used. The pre-trained models are used for the improvement of recognition accuracy, even with limited training data. Experimental results demonstrate the effectiveness of CNNbased facial recognition models in achieving high accuracy and robustness across varying conditions. This research segmented the process into three parts: Testing, training, and validation. Firstly, the proposed CNN model was trained with this dataset. Moreover, some pre-trained models are also run. They are: Inceptionv3, EfficientNet B0, EfficientNet B6, Xception, and Resnet50. This contributes to the field of facial recognition by offering a comprehensive exploration of CNN-based techniques and addressing real-world challenges.