BOUTAGHANE , Imane TADJER , Yousra2024-10-072024-10-072024http://dspace.univ-skikda.dz:4000/handle/123456789/2556In this master’s thesis, our primary focus revolves around person identification from facial images, a field deemed immensely significant across various sectors. This technology is particularly crucial in law enforcement for criminal identification, in healthcare for patient verification and personalized treatment, and in financial institutions for secure transactions and access control. The precision and efficiency of facial recognition systems hold paramount importance in averting security breaches and ensuring dependable identification processes. Harnessing the strides made in artificial intelligence, particularly in the realm of computer vision, serves to augment the efficacy of these systems. Our study embarks on an experimental journey, meticulously comparing a spectrum of pre-trained models through transfer learning, encompassing diverse architectures of Convolutional Neural Networks (CNNs) and Vision Transformer (ViT) models, with the aim of pinpointing the most optimal model for facial recognition task. Specifically, we investigate models such as EfficientNet, DenseNet, ResNet-50, VGG16, and MobileNet among the CNNs, alongside ViT-B/8, ViT-S/16, and ViT-L/16 among the ViT models. Our methodology hinges on the utilization of two pivotal datasets: The widely recognized Labeled Faces in the Wild (LFW) dataset, a staple in facial recognition research, and the Pins dataset, housing images of renowned personalities. The process of fine-tuning these pre-trained models on these datasets acts as a catalyst in optimizing their performance. The findings are immensely encouraging, with the EfficientNet model exhibiting an unparalleled classification accuracy of 100% on the LFW dataset and 94.12% on the Pins-FR dataset. These results underscore the exceptional performance prowess of EfficientNet in the field of facial recognition, eclipsing both the other CNN models and the ViT models subjected to testing.enMasters degree thesis