Research On Single-Channel Blind Deconvolution Algorithm For Ancient Chinese Character Repair

The exploration of human civilization is inseparable from archaeology,and ancient books occupy an important position.There are many reasons for the damage of cultural relics.Relying on manual identification of text information requires a huge workload,long time and low efficiency.With the continuous deepening of computer applications in archaeological research,its research methods have become more intelligent,informatized,and integrated.In actual research,the damage of the text information of ancient books and cultural relics can be regarded as a mixture of ancient Chinese character images and occlusion images,that is,only one mixed image is the researchable object,there is no prior knowledge of the mixing method,and single-channel blind source separation aims to separate the source signal from the received only mixed signal,thereby extracting useful information,and has important research value in processing images,music,and video.In the research on restoration of ancient Chinese characters,although the loss caused by the convolutional mixing method accounts for the majority,it is not the only one.Traditional blind source separation algorithms are based on the separation of mixed signals when the mixing method is known.For the problem of ancient Chinese character repair,this paper proposes a single-channel blind deconvolution algorithm based on a deep convolution generation confrontation network and a triple generation confrontation The blind source separation algorithm of the network.The DCSS algorithm faces the most convolutional hybrid methods in the research of ancient Chinese character restoration.The algorithm uses a single-channel blind deconvolution method.First,the ancient Chinese character image training set is preprocessed based on the deep convolution generation confrontation network,and then the single-channel The blind deconvolution problem is turned into a Bayesian estimation problem.Finally,based on the gradient optimization method,the estimated value is reconstructed multiple times,and the result with the smallest error is selected to separate the ancient Chinese character image,the occlusion image and the mixed matrix in the mixed image.The Tre SS algorithm is mainly aimed at the loss of images of ancient Chinese characters without mixing.The algorithm develops a novel triple generation confrontation network mechanism.Compared with the traditional single generation confrontation network,the loss function for completing the image reconstruction error in Tre GAN is repeatedly optimized through closed-loop training.The network generator,through continuous learning and optimizing the mapping relationship between the loss image and each pixel of the ancient Chinese character image,only relies on a single mixed image to separate the occlusion image and the ancient Chinese character image,so as to achieve the effect of repairing the ancient Chinese character image.The experiment is based on the ancient Chinese character data set to separate the convolutional mixed and linear mixed loss images to achieve the effect of repairing the image,and uses the peak signal-to-noise ratio,structural similarity,and Bap coefficient similarity three commonly used image quality evaluation methods Prove the effectiveness of the algorithm.