Development of a novel super-resolution method using edge-aware focal pixels

Super-resolution is a technique that aims to generate high-resolution images from low-resolution inputs and plays a critical role in domains such as image processing, computer vision, medical diagnostics, and remote sensing. By reconstructing lost details in images, super-resolution enhances visual quality and enables their effective analytical use. Today, methods used for this purpose range from traditional interpolation techniques to deep learning-based models. In this doctoral dissertation, prominent studies in the field are reviewed and common issues affecting super-resolution performance are identified. It is observed that many super-resolution algorithms fail to preserve sufficient detail around object edges. To address this limitation, a novel edge-focused algorithm is proposed along with an adaptive matrix named Diamond-Eye, which operates specifically on problematic edge pixels. Experimental evaluations conducted on standard benchmark datasets (Set5, Set14, BSD100, Urban100, DIV2K) demonstrate that the proposed method yields promising results across both pixel-based and perceptual quality metrics, including PSNR, SSIM, A-DISTS, and LPIPS. The study presents a holistic approach that targets not only visual enhancement but also perceptual quality, offering a solid foundation for future research. Another major focus of this study is to systematically examine the impact of achromatic-axis color spaces such as YCbCr and CIELab as alternatives to the RGB color space in super-resolution tasks. The findings indicate that converting super-resolved images into these alternative color spaces may result in higher scores in traditional metrics like PSNR and SSIM. However, assessments using perceptual metrics such as A-DISTS and LPIPS reveal that these score increases can be misleading and do not necessarily reflect real improvements in visual quality.