Robust Image Enhancement and Restoration for Real-World Computer Vision
This chapter presents a comprehensive study of image enhancement, restoration and preprocessing techniques for real-world computer vision applications. It begins by examining the physical sources of image degradation, including noise, blur, illumination variation, compression artifacts and adverse environmental conditions such as low light and weather effects. The limitations of vision systems trained under ideal conditions are highlighted through their degraded performance in unconstrained environments. The chapter reviews classical approaches, emphasizing their interpretability and efficiency. It then explores modern deep learning techniques such as convolutional neural networks, generative adversarial networks, transformers and diffusion models, which achieve superior performance through data-driven learning. Hybrid frameworks that integrate classical and learned methods are discussed as a practical solution for balancing robustness, interpretability and performance.