Performance Evaluation of Deep Learning-Based Facial Recognition Models on Mobile Computing Environments

Over the past few years, facial recognition technology has become an essential element in a range of applications, such as security systems, user verification, a nd s ocial media networks. The deployment of face recognition models on low-resource devices, such as low-end smartphones, poses considerable challenges owing to the restricted processing power and memory capacity at hand. The objective of this research is to determine the optimal deep learning-driven facial recognition algorithm that can provide prompt results in resource-constrained environments. The study conducted a comparative analysis of four widely used models, namely FaceNet, OpenFace, DeepFace, and VGGFace, with respect to their performance on the LFW (Labeled Faces in the Wild) dataset, which comprises a range of facial images. The assessment criteria comprised of accuracy, recall, F1 score, inference time, and memory utilization. The study employed an extensive codebase to preprocess and generate facial embeddings for each model, assess their efficacy, and present the findings in a visual format. The results of t he study indicate that FaceNet exhibited superior performance compared to the other models, achieving an accuracy rate of 94.04 %. Additionally, FaceNet demonstrated the shortest inference time of 0.0181 and the lowest memory usage. The exceptional proficiency exhibited by FaceNet renders it a highly suitable option for conducting facial recognition in realtime on devices with limited resources. The present investigation underscores the significance of carefully choosing suitable facial recognition models that are tailored to specific c ontexts a nd l ays t he g roundwork for forthcoming studies aimed at enhancing efficacy i n restricted settings.