Self-Supervised Deep Learning for Blind Image Super-Resolution: Enhancing Image Clarity

Blind Image Super-Resolution (BISR) aims to reconstruct high-resolution (HR) images from degraded low-resolution (LR) inputs without prior knowledge of degradation types. Traditional approaches, including bicubic interpolation and supervised deep learning models, rely on paired LR-HR datasets, limiting their applicability to real-world scenarios with unknown degradations. In this work, we propose a self-supervised deep learning framework based on Super-Resolution Convolutional Neural Network (SRCNN) to enhance image clarity without external HR supervision. Our method generates self-supervised LR-HR pairs dynamically, allowing SRCNN to learn robust feature mappings independent of predefined degradation models. We evaluate our approach on benchmark datasets and demonstrate significant improvements over traditional interpolation and supervised deep learning methods in terms of Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM). The proposed framework reduces dependence on labeled data while improving generalization, making it suitable for applications in medical imaging, satellite imagery, and digital forensics.