Research On Image Style Transfer Method

Image style transfer is to transform the style of an image to a natural image,so that the original natural image retains the original content and has a unique artistic style.The traditional style transfer algorithm realizes the style transfer of the image by manually analyzing the image style and designing a specific mathematical model.The disadvantage is that it can only extract the underlying features of the image,and can only simulate a specific style.With the rapid development of artificial intelligence,image style transfer technology uses deep learning technology to achieve rapid and arbitrary style image style transfer.In order to improve the quality of stylized images,based on the research of existing image style transfer technology,this paper proposes an image style transfer method based on edge preservation and an image style transfer method based on image aesthetic quality score.In order to solve the problem of edge deformation of stylized images,this paper proposes an image style transfer method based on edge preservation.First,using the advantages of morphology in edge detection,a new edge detection method based on morphology is designed;Secondly,the edge detection method is used as the edge detection module to combine the image generation network based on the encoder-decoder structure and the VGG network to build an image style transfer network based on edge preservation;then,an edge loss function is designed according to the output result of the edge detection module;the edge loss function,content loss function,and style loss function form a total loss function to train a style transfer network based on edge preservation.After experimental verification,the image style transfer based on edge preservation proposed in this paper can make the edge information of the generated stylized image and the edge information of the content image as consistent as possible,and improve the structure preservation of the stylized image.In order to enhance the visual appeal of stylized images,this paper proposes an image style transfer method based on image aesthetic quality score.The image aesthetic quality score uses a computer to analyze the aesthetic factors of the image,simulates the human perception of the image,and scores the beauty of the image.This paper combines the image aesthetic quality scoring network with the image generation network based on the encoder-decoder structure and the VGG network to build an image style transfer network based on the image aesthetic quality score;and according to the output of the image aesthetic quality scoring network,an image aesthetic quality loss function is designed;the image aesthetic quality loss function,the content loss function,and the style loss function form a total loss function to train a style transfer network based on the image aesthetic quality score.After experimental verification,the image style transfer method based on image aesthetic quality scoring proposed in this paper can meet people's different needs for image style transfer,laying a foundation for the expansion of the application range of style transfer.