<![CDATA[This paper provides a review of previous studies in Low-Resolution Face Recognition (LRFR), specifically focusing on cross-resolution Face Recognition (FR) methods. While state-of-the-art deep learning FR systems achieve high accuracy on high-resolution (HR) images, they are generally unsuitable for low-resolution (LR) images frequently encountered in applications like surveillance systems, where faces often have low pixel counts due to capture conditions. Cross-resolution FR, which compares an HR image with an LR image, presents a significant challenge due to the distinct visual properties of images at different resolutions. The paper discusses two primary approaches to address this problem: super-resolution (SR), which is a transformative method that aims to construct HR images from LR ones, and unified feature space (UFS), a non-transformative method that maps facial features from varying resolutions into a shared feature space. This work summarizes both SR and UFS methods. Based on the review, the paper concludes that non-transformative (UFS) methods are more suitable for future directions. This recommendation is driven by their lower computational power requirements, proven effectiveness in real-world implementations such as mobile devices and drones, and alignment with current technological trends. The paper also emphasizes the need for further research using real or natural LR face images to identify degradation patterns and compare results between real and artificially generated LR images.]]>
Copyrights © 2025
