Robust Image Processing Techniques for Complex Real-World Environments
Image processing systems that are designed to work in real environments are subject to factors such as haze, rain, noise in sensors, motion blur, lack of proper illumination, and compression. Unlike the controlled environment, the aforementioned factors require an approach that can withstand various levels of degradations and still produce reliable results. In this chapter, a comprehensive discussion of how image processing systems have evolved, starting from basic signal-processing techniques to current advanced AI-based ones like CNNs, vision transformers, GANs, Retinex Enhancement Models, and Diffusion models is presented. Applications of these methods in autonomous vehicles, medicine, remote sensing, industry, and security surveillance are considered. Key issues such as computational complexity, dataset size, domain adaptation, and adversarial attacks are explored.