Research On Color Restoration Of Mural Image Based On Convolutional Neural Network

Mural is one of the most important cultural relics in human history.These murals truly reflect the important information of religion,culture,art,humanities,social life,natural forms and other aspects,and have extremely important protection and research value.However,after hundreds of years of milling and human influence,the colorful colors on many precious murals have long faded.The problem of fading and discoloration of murals has seriously affected the development of the research and display of mural images.In recent years,the virtual color restoration of mural image based on image processing technology has gradually become a research hot spot in the field of mural image virtual restoration.The virtual color restoration of image is to restore an image with color degradation to a state before it is not degraded through a series of restoration processes.The method based on deep learning is an important direction of the research on the virtual color restoration of image.In this paper,based on the virtual color restoration method of image based on convolutional neural network,the mural image is taken as the object,and the virtual restoration method of mural image color is researched.Details as follows:1.Proposed a method of virtual color restoration of mural image based on multiple constrained convolutional neural network.In view of the fact that the color restoration method based on traditional mathematical statistical information is not good for the mural image with a wide range of colors,it may cause the problem that the color of the mural image after color restoration is not restored or there are false colors and artifacts.This paper uses a convolutional neural network extract the high-level features of the mural image;uses the maximum mean difference constraint to establish a global feature mapping relationship,extract the color features of the mural image and retain the global position information;uses markov random field constraints to restrict the spatial layout of the mural image,retain the mural image Local color and structure information.This method can make full use of the color information of the mural image,solve the problem of missing color information in the extraction of only a single color feature,and improve the accuracy of color restoration.Experimental analysis shows that the method in this paper has a good effect on the color restoration of the faded area of the mural,and can better overcome the color transition distortion and improve the quality of the virtual restoration of the color of the mural image.2.Proposed a method of virtual color restoration of mural image combining semantic segmentation and convolutional neural network.In view of the characteristics of many non-local self-similar image blocks such as texture,color and structure in the mural image,aswell as the problem of color mismatch in the color restoration area of the mural image and poor restoration of the location details,this paper proposes a mural image combined with semantic segmentation color restoration method.This method is based on the semantic segmentation technology of dilated convolutional neural network to semantically segment the local pixel blocks of the mural image,so that during the color restoration process of the mural image,the problem of mismatching the content of the color restoration area is avoided;the segmentation is performed using the convolutional neural network After the results are extracted,the accuracy of color restoration is improved;the noise during the color restoration of the mural image is suppressed by the square gradient constraint.Through the experimental analysis of the mural image,it is shown that the method of this paper can effectively improve the accuracy of the virtual restoration of the color of the mural image and the fidelity of the color restoration of the mural image.