Face recognition using threshold string representation matching under unconstrained conditions

This paper introduces a new approach called the threshold string representation matching (TSRM) algorithm, designed to enhance face identification and verification in scenarios by partial occlusions like sunglasses, scarves, and caps. The TSRM algorithm innovatively manipulates pixel features by calculating the difference between the centre pixel and its neighbours rather than relying on the traditional feature compression method (CSM-RL), where pixel difference is taken row-wise. This technique also improves the robustness of facial expression recognition by focusing on localised spatial relationships within the image patches. Our algorithm provides a novel feature-matching algorithm which effectively matches the prob image with the gallery image in the occlusion condition. We evaluated our proposed TSRM approach on two public face datasets: the AR dataset, which includes natural occlusion, and the ORL dataset, for synthetic occlusion.