Comprehensive survey on deoldifying images and videos

Abstract

In the field of image processing, Deoldify refers to the revitalization of aging visual media, such as historical photos and videos, which present unique challenges due to accumulated defects, unpredictable degradation, and physical damage. Traditional restoration techniques, such as manual retouching, chemical treatments, and manual colorization, are often insufficient for addressing the complexity of these tasks, particularly in areas such as denoising, super-resolution, brightness enhancement, deblurring, colorization, compression, and inpainting. These methods lack automation, scalability, and precision, especially when dealing with severely degraded media. This survey highlights the limitations of conventional approaches. We focus on how recent advancements in deep learning, including convolutional neural networks, variational autoencoders, generative adversarial networks, Transformers, and diffusion models, have surpassed traditional methods in these subtasks. By leveraging deep learning, tasks such as noise reduction, contrast restoration, and resolution enhancement are performed with greater accuracy and efficiency, significantly improving restoration outcomes. This work aims to provide a comprehensive review of these techniques, showcasing their superiority over traditional methods while identifying challenges such as dataset limitations and the need for better handling of extreme degradation,and proposing directions for future research in old media restoration.