Research On Colorization Method Of Gray-scale Image Of Farmer’s Painting Based On Generative Adversarial Network

Qinghai farmer’s painting originated in the 1970 s.Qinghai farmer’s painting is a unique kind of painting in traditional folk art in Qinghai.Most of the relevant images recorded in some ancient documents and publications are gray-scale images,which are not conducive to the research and digital protection of Qinghai farmer’s painting.Therefore,it is very necessary to use the gray-scale image colorization technology to realize the colorization of gray-scale image of Qinghai farmer’s painting.This paper deeply studies the application of generative adversarial network in the colorization of gray-scale image of Qinghai farmer’s painting.The main work of this paper includes the following parts:（1）Construction of Qinghai farmer’s painting image dataset.At present,the gray-scale image colorization methods are mainly aimed at natural images,and there is no image dataset related to Qinghai farmer’s painting.Therefore,this paper carries out post-processing and screening on the currently collected Qinghai farmer’s painting images,extracts the characteristic patterns of the images,and obtains 9574 Qinghai farmer’s painting related images through standard processing and manual screening,so as to construct Qinghai farmer’s painting image dataset.（2）Research the traditional gray-scale image colorization methods and the traditional generative adversarial network model,and analyze their advantages and disadvantages.The traditional gray-scale image colorization methods need cumbersome manual interaction,and the colorization efficiency is low.In the process of colorization,there are also the problems of insufficient ability of image edge information processing and color leakage.Although the traditional generative adversarial network can realize the automatic colorization of the gray-scale images,it has the problems of low utilization of image feature information and insufficient processing ability of image detail information.As a result,the color effect of gray-scale image of Qinghai farmer’s painting realized by traditional generative adversarial network is not ideal.（3）Research and improve the Pix2 Pix generative adversarial network model.Although Pix2 Pix generative adversarial network improves the processing ability of image detail information and alleviates the problem of color leakage in the process of colorization,the color contrast of the generated color image is low,resulting in the dark visual effect of the generated image.To solve the above problems,this paper improves the Pix2 Pix generative adversarial network,uses Leaky Re LU as the activation function of the network,and uses the convolution layers to replace the maximum pool layers of the original generative network,so as to retain more image feature information.The improved network can restore more image color details and alleviate the problem of low color contrast of the color images generated by the original network.The experimental results show that using this method to realize the colorization of the gray-scale image of Qinghai farmer’s painting can obtain more ideal and higher quality color images.（4）Design and implement the Qinghai farmer’s painting gray-scale image colorization system.According to the demand and function analysis of the system,build Qinghai farmer’s painting gray-scale image colorization system,provide the visual interface to complete the colorization of gray-scale image of Qinghai farmer’s painting,and generate high-quality Qinghai farmer’s painting color image.