Research On Murals Inpainting Based On Generalized Regression Neural Network

In this paper,the digital repair technology was adopted for the restoring treatment of the damaged images in the Dunhuang frescoes.First,the background and significance of the restoration of the images in the Dunhuang frescoes and the related research status at home and abroad were summarized,the content of each chapter in this paper was briefly introduced,the relevant theories and methods of digital repair technology were outlined and the evaluation standards for image restoration were presented.In the third chapter,the concept,classification and characteristics of the neural network were explored,and the adoption of the Generalized Regression Neural Network(GRNN)in the artificial neural network was put forward for the researches on the digital restoration and protection of the Dunhuang frescoes.This paper was designed to study how to improve the accuracy and efficiency of image restoration using the generalized regression neural network.The research mainly covered the following aspects:Firstly,the experimental analysis of the feasibility of applying the generalized regression neural network to the image restoration was conducted.The results showed that the generalized regression neural network had simpler structure,faster processing speed and more accurate precision,hence verifying the validity of this method.Secondly,the image preprocessing was executed.First,the noise of the mural image to be restored was removed using the anisotropic diffusion in order to keep the image edge;and then the morphological dilation operator was used to extract the boundary pixels in the region to be restored to make preparation for the image restoration.Finally,the sample pixel block similar to the pixel of the neighborhood on the edge of the region to be restored was used as the input training sample of the generalized regression neural network,and the pixel value of the sample block was scanned and input to the generalized regression neural network using the zigzag shape.Meanwhile,the self-adaptive smoothing parameters were used,and finally the approximate generalized regression neural network restoration model was obtained.The experimental result showed that this method had a certain effect on digitalrestoration of murals.Finally,the content of this paper was summed up and the the future prospects were put forward.