Generative Artificial Intelligence for Historical Image Reconstruction Enhancing Visual Interpretation in History Education Programs (Published)
Historical Educational engagement. There is a void in learner-focused historical reconstruction since current restoration methods seldom suit educational visualization objectives. To improve visual interpretation in history education programs, this research attempts to create a generative artificial intelligence model that reconstructs faded or damaged historical material. The suggested technique employs the Historical Heritage Painting Collection dataset, having 2,192 images. Data preprocessing incorporates noise reduction using a non-local means (NLM) denoising technique, alongside contrast normalization and scaling for uniformity. To maintain original object boundaries and direct reconstruction integrity, edge-based mapping is used for structural feature extraction. The suggested Dynamic Vortex Optimizer-driven Stable Diffusion with ControlNet (DVO-SD-CN) model is then fed the processed pictures and generated structural maps. Here, ControlNet conditions Stable Diffusion on extracted structural features to guaranty that the generated content stays true to the original historical context rather than creating irrelevant details, while DVO, a vortex-inspired metaheuristic technique, adaptively adjusts the diffusion model’s hyperparameters and denoising trajectory for stable convergence. Accurate, structure-aware historical image reconstruction is made possible by this integrated architecture, making it appropriate for interactive learning modules, digital storytelling, and classroom presentation. The proposed DVO-SD-CN achieves PSNR values of 33.84, 32.47, and 29.86 for low, moderate, and high deterioration, respectively, according to experimental evaluation. All things considered, the suggested approach shows great promise for improving visual interpretation in history education programs using dependable, technologically advanced picture reconstruction techniques.
Keywords: controlnet, generative artificial intelligence, historical image reconstruction, history education, stable diffusion, visual interpretation