Akant, ZeynepGhazaei, ElmanBalcisoy, SelimCampana, StefanoFerdani, DanieleGraf, HolgerGuidi, GabrieleHegarty, ZackaryPescarin, SofiaRemondino, Fabio2025-09-052025-09-052025978-3-03868-277-6https://doi.org/10.2312/dh.20253174https://diglib.eg.org/handle/10.2312/dh20253174Colorizing historical images and modernizing traditional attire are key to bridging past and present in digital heritage preservation. Accurate colorization improves the interpretation of old photos, while modernizing historical attire supports cultural adaptation and fashion preservation. This paper presents an unsupervised method for colorizing 19th century images using GANs, trained with a dataset from modern-historical films. By leveraging the GAN discriminator, realistic colorizations are generated without paired data, capturing the textures and authenticity of historical scenes. A diverse film-based dataset enables the model to generalize across eras. Additionally, historical clothing is segmented and transferred onto modern subjects using diffusion-based virtual try-on techniques. Together, these methods support cultural preservation by blending historical accuracy with modern representation.Attribution 4.0 International LicenseCCS Concepts: Computing methodologies → Image colorization; Computer vision tasks; Generative adversarial networksComputing methodologies → Image colorizationComputer vision tasksGenerative adversarial networks10.2312/dh.202531745 pages