GAN-based Image Enhancement Algorithm for Personnel Archives Scanned Image

In the digital processing of personnel Archives, the image quality of the files directly determines the effectiveness of digitization. This paper proposes an enhancement algorithm for scanned personnel file images to improve the efficiency and quality of digital processing based on generative adversarial network(GAN). An improved Swin Transformer-based generator model is designed, which utilizes the original image and document image parsing results as inputs to the backbone network. The model incorporates a channel-spatial attention mechanism fused with the Swin Transformer module for feature extraction. Additionally, a lightweight CNN model is designed as the discriminator, and a multi-task learning loss function is adopted for training to enhance the image quality generated by the generator. Experimental results demonstrate that the proposed method outperforms in both subjective and objective evaluation metrics, achieving significant improvements in enhancing the quality of scanned personnel file images.