Research Cultural Relic Restoration System Based On Deep Learning And Super Resolution

Cultural relics are the symbol of history and an important window for people to understand history,culture and art in modern society.Therefore,the restoration of cultural relics is conducive to the protection of cultural relics,the promotion of cultural inheritance and the digital development of cultural heritage.At present,tens of millions of broken cultural relics exist in our country,and there are not enough professional restoration personnel,and the restoration relies excessively on personal experience.Therefore,science and technology is introduced in the field of restoration,on one hand,to make the restoration more efficient and accurate,and to avoid secondary damage,on the other hand,promoting the digital preservation of cultural heritage,realizing visual relics,diversified display,and promoting cultural exchange.Since there is no publicly available data set of cultural relics,this thesis takes cultural relics collected in museums as the research object,such as porcelain,bronze ware,calligraphy and painting,etc.,and studies the cultural relics images with wear,cracks and a small part of imperfections.The specific research content of this thesis is as follows:(1)Create datasets of cultural relics.The web crawler technology is used to collect cultural relics images,such as porcelain,jade,embroidery and other cultural relics images,and screen them.The images that meet the conditions are taken as the initial cultural relics data set,and the data dynamic enhancement method is used to expand the training process,so as to increase the diversity of training data and improve the generalization ability and performance of the model.(2)Design the restoration network of cultural relic images.A cultural relic restoration network is built with the framework of the antagonistic network.Firstly,a two-branch network was designed to repair edges and global textures respectively.Secondly,the feature fusion structure makes full use of the edge and global feature information,and the edge information guides the image restoration process to ensure that the repaired image will not appear obvious errors.Then,the Res NET network is designed as a network discriminator to distinguish true and false information and provide feedback signals,thus improving the performance of the generator.Finally,edge information is introduced to guide network update in the fight against loss.Through comparative study,it is found that this method has better edge effect and the overall semantic coherence of the repair results.(3)Improve the super resolution reconstruction algorithm based on generative adversarial network(GAN).In view of the current problems such as feature missing and local fuzzy distortion in the middle and high level of super resolution reconstruction,the experiment in this chapter uses SRGAN as the framework.Firstly,the residual block in SRGAN network is improved.In order to learn the high level features more fully,the multi-scale residual network is used to extract the multi-scale feature information.Secondly,the attention mechanism is introduced and the multiscale residual network is designed to enhance the adaptive learning of context multifeature information and improve the accuracy of the model.Through comparative study,it is found that this method can get higher quality clear image.(4)Design the restoration system of cultural relics image.Firstly,the functional requirements of the cultural relic system are analyzed,and then Py Qt module is used to design the cultural relic restoration and super resolution reconstruction system.