Research And Implementation Of Thang-ga Image Inpainting System Based On Deep Learning

As an important part of Xixia art,the Thang-ga image has special historical and artistic research value.In recent years,with the continuous advancement of Xixia studies and the continuous emergence of archaeological excavations,the restoration work of the ancient Thang-ga cultural relics with different degrees of damage has gradually been put on the agenda.At present,in the field of cultural relic restoration,most of them are mainly manual restoration.With the continuous expansion of the field of deep learning,more technical support is provided for image restoration.The virtual repair of Thang-ga images combined with advanced deep learning technology can not only avoid the risk of directly repairing the "secondary damage" of cultural relics,but also provide the necessary digital resources for digital display and traditional cultural communication.The main work of this paper is as follows:(1)In view of the problem that there is no publicly available Thang-ga image data set,this paper collects and preprocesses Thang-ga image through various ways to create a Thang-ga image data set;(2)Based on the obtained Thang-ga image database,this paper proposes a U-Net-based generative adversarial network repair algorithm to repair the damaged Thang-ga image.Introducing U-Net into the generator,hoping that the U-Net network can adapt to the problem of relatively rare Thang-ga images while constructing a repair network,and propose a new and improved way to repair Thang-ga images;(3)The U-Net-based generative adversarial network inpainting algorithm is improved,and a Thang-ga image inpainting algorithm based on Dilated-DCGAN is proposed.In the improved algorithm,the hollow convolution and convolutional neural network are introduced into the generator and the U-shaped structure is retained.And we reduces the number of up and down samples in the generator.By alternately optimizing the generator and discriminator,we get a network model with good repair effect.Using the Thang-ga image inpainting algorithm proposed in this paper,through the comparative analysis of the results of multiple sets of inpainting experiments,it is verified that the algorithm has a good repair effect;(4)Use the Thang-ga image inpainting algorithm proposed in this paper to design and build a Thang-ga image inpainting system to realize the repair of damaged Thang-ga images,and to browse and query different types of Thang-ga images.