Research On The Repair Algorithm Of Irregular Interference In Ancient Stone Carving Documents

Ancient stone inscriptions have always been an indispensable research content in the study of history and culture in my country,and have important historical material value.However,due to the erosion of the natural environment or manmade destruction,the surface of the stone inscriptions has appeared in a number of different sizes,The unevenly distributed and changeable shape of the interference area not only affects people's perception needs,but even hinders historical research.With the advent of the information age,important stone carvings can be stored digitally to extend the preservation time of cultural relics,and stone carvings can also be shared through the network to break the restrictions of time and place,and increase the audience of ancient history and culture.,Use positive and positive guidance methods to promote the spread of Chinese excellent culture.Faced with the restoration of a large number of stone inscriptions,manual restoration is inefficient and easy to cause secondary damage.To sum up,this article is a useful exploration of digital restoration of damaged stone inscriptions.On the basis of saving manpower and material resources,as much as possible Repairing based on the effective pixels in the stone carving document image will not only promote the digital protection of the stone carving document to a certain extent,but also avoid damage again.The research content of this article is as follows:First of all,discuss the lack of existing text data sets,and propose methods for establishing irregularly disturbing stone carving literature databases.At present,there is no complete public data set of stone carving documents,which has caused troubles to the problem of irregular interference repair of stone carving documents.In this paper,by analyzing the natural damage of stone carving documents and the data characteristics of damaged patterns,the stone carving document samples in real scenes are collected to construct a text sample data set.After image preprocessing,this paper proposes an interference pattern detection algorithm to detect damaged stone carving documents.The damaged area in the image is collected for data,and randomly cut,rotated,and folded during training to further expand the sample types of the interference Mask database.In the next part,construct a model for repairing irregular interference of stone inscriptions.Existing text image restoration research focuses on line-type interference of different colors,and cannot solve the problem of text image interference in real scenes.In response to this problem,this paper uses the U-NET framework to restore the damaged stone inscriptions,and effectively predicts the missing areas through partial convolution.Under the common constraints of perceptual loss,style loss and perceptual loss,an irregular restoration model of stone inscriptions is constructed.by reconstructing the missing the segment according to the prior font style,a breakthrough shape,color limits stone Document image interference pattern.Different from natural scene images,text images have extremely high requirements for the internal structure of text,that is,the detail restoration of the reconstructed image area.This article uses the text recognition interface to indirectly verify the effectiveness of text restoration to ensure that the restoration details are reasonable.Finally,repair the pictures of real stone inscriptions.Due to the influence of scanning equipment,light,preservation environment,etc.,the original images of stone inscriptions in the real scene often cannot reach the ideal state of restoration.They need to be restored through a series of processing such as tilt correction,grayscale and binarization Original Picture.After preprocessing,the greedy algorithm is used to calibrate the irregularly disturbed stone inscription image interference area,and the pre-trained model is used to gradually repair the effective pixels around the defect area.Experiments show that the model can effectively predict the missing part of the text based on the font,shape and stroke direction of the test picture.The highest PSNR reaches 32.58 d B,and the best LOSS reaches 0.015.Compared with other restoration methods,this method not only solves the problem of the same color irregular interference restoration,but also significantly improves the subjective visual effects and image quality evaluation results in the restoration of text images,and the text recognition rate increases by 30.49% before and after restoration.In the restoration of stone carving documents in real scenes,the interference area is effectively calibrated,and text with rich details and smooth stroke edges can be reconstructed according to the font shape of the test picture,which greatly reduces the effort and time required for artificial restoration.