Edge-based image super-resolution using generative adversarial network for iris detection in very low-quality images

Image super-resolution (SR) techniques have seen significant advancements in recent years, particularly with the use of deep convolutional networks. However, challenges remain in some cases like preserving edge details. This paper presents an enhanced approach to single image super-resolution (SISR) by introducing a new perceptual loss component, Edge Energy Loss, which addresses the issue of overly smoothed super-resolved images. The proposed method, Edge-SRGAN, integrates this loss with the SRGAN framework to improve the detail and perceptual quality of edges in super-resolved images. Specifically, this approach is applied to eye image super-resolution, where precise edge representation is critical for accurate iris detection. The Edge-SRGAN method significantly enhances iris detection accuracy in low-resolution eye images, achieving a substantial improvement from 28% to 94% accuracy without relying on complex iris detection algorithms. Validation through Mean Opinion Score (MOS), Peak Signal-to-Noise Ratio (PSNR), and Structural Similarity Index (SSIM) metrics confirms the effectiveness of the proposed approach. Additionally, a natural low-resolution dataset is generated using deep networks to further refine the training and performance of the Edge-SRGAN.