Research On Application Of Image Restoration Technology Of Ancient Books Based On Adversarial Generative Network

Ancient book is a bridge for the spread of human culture.It is an important way to understand the production and life of ancient people.Research on image restoration of ancient documents will help promote the development of local tourism,economy,culture,and education.At the same time,it also has an impact on ancient history and culture.The construction of spiritual civilization is of great significance.However,at present,the research on the digital restoration of damaged images of ancient books is still in the initial stage of research.A large number of ancient books are damaged due to environmental corrosion,paper wear and lack of protection.At the same time,due to the limited ability of the hardware of the image acquisition equipment,the digitized images of some ancient books are low in resolution,causing inconvenience for the later image processing and research work.Therefore,digital image restoration technology can be used to solve image damage and low image resolution.Generative Adversarial Networks(GANs)in Deep Learning is widely used in the field of image generation due to its superiority in network structure.However,the confrontation generation network still has the following three problems in the process of digital image restoration in ancient documents:1.The problem of generating textures for specific regions in the image of incomplete ancient books still needs to be solved;2.Due to the limitation of the resolution of the acquisition equipment,low resolution The image enhances the resolution and image details.3.Constructing a stable network structure can quickly create problems against the generation model.In this paper,we use the method of confrontation generation network to study the restoration technology of digital images of ancient books.The specific work content is as follows:(1)A counterattack generation network algorithm model for fixed area texturerepair is proposed.In this paper,the local loss function is used to improve the additional information confrontation generation network,and the network structure of the additional information confrontation generation network is optimized.The loss function of the model generation based on the combination of the local loss function and the global smooth loss function is proposed,and the incomplete texture of the fixed region of the ancient document is generated.(2)Propose a super-resolution algorithm model for low-resolution image countermeasure generation network.This paper constructs the confrontation generation network model of Laplacian pyramid structure.This model takes multi-layer image pyramid reconstruction method into account,taking into account the macroscopic style of the image and the high-frequency details of the image.The noise is used as the input layer of the network model to simulate the high-frequency distribution of real data and generate high-frequency details.Based on the sampling of the original image,high-resolution images with richer detail effects are generated.(3)A noise reduction algorithm based on the network loss function of confrontation is proposed.The noise image is discriminated by the isolated pixel detection method for reducing the noise in the process of generating an image against the generation network.Thanks to the network structure of the isolated pixel detection method,noise of 60%of the image width can be reduced.(4)Established a digital image restoration prototype system for ancient books.Corrupted,low-resolution images of ancient books are given corresponding repair images,providing researchers of ancient documents with a way to give fast and objective repair images.In summary,the network structure of the confrontation-generating network mentioned in this paper can be used for image restoration of digital image incomplete and low-resolution problems,and can be applied to a wider range of image restoration problems.At the same time,according to the experimental results in this paper,the human eye recognition error rate in image restoration technology reaches 83%,and super-resolution technology achieves 4 times the image pixel expansion.