Face recognition is a critical and challenging task in the field of computer vision and biometrics, with
applications ranging from security and surveillance to human-computer interaction. In recent years, deep
learning techniques, particularly deep convolutional neural networks (CNNs), have revolutionized the face
recognition landscape by achieving remarkable accuracy and robustness. It is a means of enhancing the quality
of Object recognition; it plays a crucial role in various computer vision applications, such as autonomous
driving, surveillance systems, and image retrieval. Deep Convolutional Neural Networks (CNNs) have
emerged as a powerful approach for achieving state-of-the-art results in object recognition tasks. The core of
this paper details the architecture and design of the proposed deep CNN-based face recognition system. It
covers key components such as data preprocessing, network architecture selection, training strategies, and
optimization techniques. An evaluation of various CNN architectures, including FaceNet, VGGNet,
FACENet, ResNet, and EfficientNet, is presented along with their respective strengths and weaknesses in the
context of face recognition. The research was aimed at creating an effective system that can accurately
recognize student faces and display corresponding profile from the database before examination. FaceNet is
the preferred CNN architecture for this work as it has comparatively better accuracy than other CNN
architectures. It explores the challenges posed by variations in pose, illumination, and occlusion, and discusses
techniques employed to mitigate these challenges within the proposed system. Additionally, the paper
included fine-tuning strategies for adapting a pre-trained model to specific face recognition tasks. In creating
this hardware, the model was pre-trained with student faces for recognition. 50 individuals, each with 20
images, were imputed as the dataset, giving a total of 1000 images in the dataset. These 1000 images were
used to test the various popular CNN models given above. The test resulted in: VGGNet having an accuracy
of 92%, precision of 90%, recall of 91%, and inference speed of 0.5: ResNet 50 having an accuracy of 93%,
precision of 91%, recall of 92%, and inference speed of 0.3: MobileNet having an accuracy of 88, precision
of 84, recall of 86, and inference speed of 0.2: FaceNet having an accuracy of 94%, precision of 92%, recall
of 93, % and inference speed of 0.4. The resulted showed that optimized models designed for edge AI
microprocessors significantly enhanced inference speed while maintaining similar levels of accuracy, as
demonstrated by the FaceNet model.
Browse additional articles authored by one or more contributors to this paper.