Research On Chinese Print Style Transfer Algorithm Based On Neural Network

Gatys et al.segmented and recombined the content and style of images for the first time and established a neural network-based image style transfer method on this basis.With the rapid development of the Internet,there are more and more researches on image style transfer technology,but the application results in the printmaking field in China are not ideal.The main reason lies in the distinctive description of knife flavor and wood flavor in Chinese prints,and most of the prints are expressed in the form of lines,and at the same time,such lines also have the overall unity.In order to deal with the style transfer of Chinese printmaking images more effectively,this paper studies the method of Chinese printmaking style transfer based on neural network.At present,the application and research of image processing of Chinese printmaking are relatively few.If the general style transfer algorithm is applied to Chinese prints directly,the migrated target image cannot achieve the expected result.Therefore,according to the artistic characteristics of Chinese printmaking,this paper proposes a relatively complete modification scheme based on the original mode of image style transfer,that is,extracting some characteristics of Chinese printmaking,recombining them through artificial neural network,and finally generating Chinese printmaking style image through generator.Main research contents of this paper:(1)Chinese print image style transfer based on improved convolutional neural network.On the basis of the structure of VGG-19 neural network model,the unique lines,textures and frequencies of Chinese prints are used as constraints to supervise the generation of images,and Total Variation Loss is added to constrain the noise of synthesized images.Then,the transfer of Chinese printmaking style is completed in an experimental way,and the results of the experiment are discussed and studied.(2)Chinese print image style transfer based on improved generative adversarial network.Firstly,the core principle of generative adversarial network model GAN is analyzed,and the relativistic discriminator technology is described in detail.Then,according to the principle of generative adversarial network,a new generator based on Cycle GAN is designed and relativistic discriminator technology is used to reduce generative adversarial loss.Finally,a qualitative analysis experiment is designed to evaluate and analyze the model.