Antique Photo Restoration and Colorization via Generative Model

In the past, many photographs of famous historical figures and moments were captured in back and white photos. Those captures are often distorted by the limitation of the old-style camera and the negative influence of the poor storing environment. It is obvious that the restoration and colorization of those images can make history lively. Since manually retouching images is time-consuming and hard to be done by people without aesthetic senses, many researchers have proposed models that automatically remove the artifacts in the old photos. However, these methods only solve either image restoration or colorization tasks which cannot fully address the task of image retouching. Consequently, in this work, we propose an effective end-to-end framework, named AIRC, for image retouching. Besides, previous works often use synthesized old photos for training but these pseudo datasets can not replicate exactly the real antique photo and prevent the trained model from being used in reality. To this end, we also introduce a new antique synthetic dataset, namely OldifiedScenes, that resembles real old photos by blending with paper and artifact textures. Quantitative and qualitative results are provided to demonstrate the effectiveness of our proposed method.