AI-BASED RESTORATION OF DIGITAL HERITAGE ARTWORKS

Digitization of digital heritage artworks is of high importance to preserve cultural memory and provide access to historically important visual objects in the long run. The digitization of heritage materials is often affected by noise, fading, physical damage and gaps in the materials, all brought about by the physical decay of original materials and shortcomings in digitization processes by heritage institutions. The current paper presents a culturally sensitive AI-based restoration system that improves the visual quality and does not affect the stylistic integrity and historical authenticity. The model combines the use of convolutional neural networks to extract local features, generative models to reconstruct and restore severely damaged or missing areas and model global context using transformers to achieve compositional coherence. Cultural limitations and expert-in-the-loop validation is added to overcome ethical and authenticity issues so as to minimize the chances of over-restoration and style bashing. The proposed technique is tested on a variety of digital heritage collections such as paintings, manuscripts, murals, folk art and archival photographs. The results of experiments indicate that in comparison with traditional and baseline AI restoration methods, the experimental methods show consistent improvements in PSNR, SSIM, and LPIPS. Case analysis and expert evaluation also affirm that the framework offers an appropriate balance between visual addition and cultural conformity, which makes it appropriate to scalable and responsible preservation of digital heritage.