Research On Heritage Images Based On Enhanced Super-Resolution Generative Adversarial Networks

The current protection and inheritance of cultural relics digitization is facing serious challenges,and a large number of low-resolution captured images of cultural relics constrain the restoration and reproduction of cultural relics digitization With the intelligence of digital technology,image content enhancement has received extensive attention from researchers.However,as a technology of image content enhancement,the research of image super-resolution reconstruction technology applied to cultural relics images is relatively rare;in addition,the existing image super-resolution reconstruction technology,in terms of image restoration,there is a lack of high-frequency detail and texture detail smoothing and other problems,and is not applicable to deal with such complex structure and colorful images as cultural relics murals.Therefore,this paper proposes an improved image super-resolution model based on the generative adversarial network framework with the goal of improving the perceptual quality of images,so that problems such as color block blurring of cultural relics murals can be effectively solved.The main work of this paper is as follows:(1)Firstly,this paper conducts research on image degradation model.In particular,in the image super-resolution algorithm based on generative adversarial network,the module of cultural relics image degradation is added,which combines the steps of image rendering degradation and noise processing to produce low,resolution sample images that are more suitable for the degradation of cultural relics mural paintings,which are used for the training of the model;in addition,the module can deal with the non-cultural relics images of 2K clarity and the cultural relics images of 1K clarity,so as to effectively expand the training set;the training set can be used for the subsequent model improvement work.The training set can be used for subsequent model improvement.(2)Then,according to the practical application requirements of cultural relics mural images,combined with the characteristics of ancient Chinese mural paintings with many details and rich colors,this paper proposes and implements a kind of improved image super-resolution model;the model uses a new network architecture,including the image texture module and the generative adversarial network,etc.,which can effectively solve the problem of the blurring of the color block and achieve the goal of retaining the details and colors of cultural relics mural paintings.Meanwhile,the performance of the improved model is compared with the original model through simulation;in terms of the Natural Image Quality Evaluator(NIQE)metrics,through the experimental comparisons on the cultural relics mural paintings datasets sy300,Ih100,and lysg val,the NIQE values corresponding to the improved model are reduced respectively by 0.40%,0.21%and 0.43%,which indicates that the model has the ability to reduce the distortion of the image and retain the texture details of the image,and better maintain the visual perception of the image.In addition,this paper visualizes the proposed model with the reconstructed images based on the super-resolution generative adversarial model,enhanced super-resolution generative adversarial model,enhanced depth super-resolution model and authenticity-enhanced super-resolution generative adversarial model,and the comparison results also intuitively illustrate that the improved model is able to effectively solve the problem of mural color block blurring,and robustness experiments have been done.(3)Finally,this paper applies the image super-resolution model in the cultural big data platform,designs the corresponding interactive interface for the user,facilitates the user to use the model contained in the interface to super-resolve the image,and successfully adds the function of super-resolution reconstruction of the platform.Among them,the user interaction function includes image uploading,image processing and image downloading.In addition,this function was also tested in this project.