Research On Image Inpainting Of Ancient Paintings Based On Optimized Generative Adversarial Networks

As the historical witness and precious remains of human history,ancient painting has rich scientific and cultural connotation and historical research significance.However,due to natural factors and man-made damage,most of them suffer from diseases such as fading,damage and mildew.The low-quality ancient painting greatly reduces the appreciation experience and seriously hinders the cultural inheritance.Therefore,it is urgent to repair them.Traditional manual repair is not only time-consuming and laborious,facing the shortage of professional technicians and other problems,but also has the risk of irreversible secondary damage.In order to solve this problem,this paper uses advanced technologies such as deep learning to carry out the research of ancient painting image restoration.Considering that the existing image inpainting algorithms have some limitations in image modeling and network structure design,and their inpainting results in ancient painting images are also slightly inferior.According to the current situation and characteristics of ancient painting images,this paper will focus on how to improve the performance of image inpainting and ensure the efficiency of network optimization.By introducing new technologies such as gated convolution and attention mechanism,combined with the solution of image texture and structure hybrid encoding,an ancient painting image inpainting algorithm based on optimized generative adversarial networks is proposed.After qualitative and quantitative evaluation and analysis,this method can usually show excellent performance better than the existing deep inpainting algorithms.The research contents and main contributions of this paper are reflected in the following two aspects:(1)In order to make up for the shortcomings of mainstream inpainting methods in image modeling,this paper proposes an information complementarity strategy,which uses a parallel and complementary way to model image texture and structural features.Then,combined with the characteristics of encoder-decoder and single-scale deep convolution networks,a two-stage image inpainting method based on information complementary generative adversarial networks is constructed,and experiments are carried out on common public data sets.The verification shows that this method can effectively restore the image structure and texture and realize high-quality progressive inpainting by successively carrying out multi-scale feature extraction and pixel-level detail filling on the damaged image.(2)In view of the single coupling mode of two-stage inpainting network and the limitations of information complementarity strategy,attention mechanism and soft gating are introduced and cross-stage information correction unit and adaptive feature fusion unit are proposed successively.An image inpainting method based on multiple attention generative adversarial networks is finally constructed.In addition,in order to improve the performance of this method in the task of ancient painting image inpainting,an ancient murals image data set is made and used for transfer learning.A large number of ablation and comparative experiments show that these two new functional units effectively improve the inpainting performance of the method.The trained optimal model is not only superior to the current mainstream inpainting methods in subjective and objective evaluation,but also can get clearer structure and more accurate color results when restoring ancient painting images.