Low-Resolution Image Enhancement using Generative Adversarial Networks
Melvin Ajuluchukwu, Atef Mohamed Shalan, Lei Chen, Yiming Ji, Emmanuel O. Balogun · 2024
Enhancing low-resolution (LR) images is crucial in the field of machine vision science. Improving the quality of LR images captured by security cameras is indispensable for forensic analysis and identification in surveillance applications. Generative Adversarial Networks (GANs) have emerged as a powerful deep-learning technique to address the super-resolution (SR) challenge inherent in long-range surveillance photos. The primary aim of this study is to enhance low-quality images of highway security surveillance using GANs architecture to optimize the image quality by generating high-resolution (HR) equivalents of the real HR images. The desired images were reconstructed using a dataset and employing Red-Green-Blue (RGB) guided thermal SR-GANs and GANs, along with perceptual loss function techniques. These approaches enhance image quality while preserving important LR image features, thereby reducing blur and glitches associated with image upscaling.