Improving Occluded Facial Recognition Accuracy by Integrating Large Language Model

Traditional facial recognition models have extremely high accuracy in routine tasks, but in the face of facial occlusion, due to insufficient recognition accuracy, they cannot distinguish important features, and some recognition systems do not perform well. In response to this issue, this article explores the integration of open-source Large Language Models into traditional neural network-based facial recognition systems to improve their recognition accuracy, especially in occluded scenes. We propose a new framework that utilizes the Lightweight Language-Vision Alignment model to enhance feature extraction and contextual understanding in occlusion face recognition tasks. The experimental results show that the recognition accuracy has been improved to a certain extent, demonstrating the potential of LLM in computer vision applications.