Low-resolution face recognition (LRFR) remains a challenging task due to the loss of critical facial details and the domain gap between high-resolution (HR) and low-resolution (LR) images. In this paper, we propose a novel architecture based on a Bi-directional Gated Recurrent Neural Network with an Attention Mechanism (BiGRU-AM) to address these challenges. The model was trained and validated on two large-scale datasets: the QMUL-SurvFace LR dataset and the CMU Multi-PIE HR dataset. Testing was conducted exclusively on LR images to evaluate the model's performance. The BiGRU-AM architecture effectively captures both local and global dependencies in facial features, leveraging the attention mechanism to enhance critical feature representation. Bayesian Optimization is further used to effectively find optimal configuration for faster convergence due to intelligent exploration of the hyperparameter space. The model achieved a testing accuracy of 91.32% and 87.2 % when trained on the LR and HR dataset respectively outperforming state-of-the-art algorithms like mpd CNN, DCR, RI-DA, Parallel-DL, ARDAA, DeIT in both scenarios. These results demonstrate the robustness and effectiveness of the proposed architecture, highlighting its potential for real-world applications in surveillance and low-resolution face recognition systems.
