Historical Building Image Restoration and Style Transfer Based on Generative Adversarial Networks

Historical building images often suffer from damage, noise, and stylistic inconsistencies during acquisition and preservation due to aging, weathering, and limitations of acquisition equipment, leading to lost structural information and degraded visual features. This paper proposes a historical building image restoration and style transfer method based on Generative Adversarial Network (GAN). First, an encoder-decoder structure is used to extract architectural texture and contour features, and a feature matching module is used to achieve global consistency of structural regions. Second, residual blocks and attention mechanisms are embedded in the generator to enhance the ability to generate local details, and multi-objective constraints such as adversarial loss, perceptual loss, and style consistency loss are introduced to optimize the generation effect. Finally, a discriminator is used to evaluate the authenticity and style similarity of the restored image, achieving a joint improvement in style transfer and structural restoration. Experiments were conducted on Historic-Building100 and HeritageDataSet. The results show that the proposed method achieves an average peak signal-to-noise ratio (PSNR) of 29.9 dB, a maximum structural similarity index (SSIM) of 0.918, and an average style matching accuracy of 94.5 %. It outperforms existing comparative models in terms of detail restoration and region fusion.