Old Damaged Photo Recovery with Style Transfer-Based Data Augmentation

Restoring damaged old photos to their original state can enhance the viewing and comprehension of the historical context and time period captured by these photos. Recent studies attempted to restore these photos using proposed methods, but the results often suffer from blurring and loss of details. Therefore, we propose a restoration model designed explicitly for damaged old photos in this paper. Furthermore, due to the limited quantity of damaged old photos compared to other task datasets, we employ data augmentation techniques to increase the quantity and diversity of training data for the restoration method. The experimental results demonstrate that our restoration method surpasses existing methods across various evaluation metrics and provides a satisfying viewing experience.