Research On Partial Clothing Image Style Transfer Based On Deep Learning

Clothing image style transfer is to convert the style into the clothing content image,to keep the original structure and outline shape of the clothing unchanged,and show the transferred style elements(such as texture,color,brightness,material,pattern).The style transfer of the entire clothing image will generally lead to distortion and deformation of the texture of the transferred image,color changes,and blurred outlines.In order to obtain more refined clothing textures,clear clothing outlines and provide personalized clothing style customization,this paper expands a study with important research significance and practical application value: local clothing image style transfer based on deep learning.In this paper,three innovative algorithms are proposed to realize the style transfer of local clothing images,and the following research work has been completed:(1)For the clothing images with complex backgrounds,it is difficult to control the style transfer in local areas of the images,and boundary artifacts are easily generated in this process.To address the problem,a CycleGAN-based method using attention mechanism is proposed for local style transfer of clothing images.The method employs VGGl6 to extract the content features and style features from the clothing images separately,and the features are subsequently input into the CycleGAN generator based on the attention mechanism.The attention mechanism is used to distribute the probability distribution information to each clothing area of the complex background to obtain the area with more attention distribution and the area with higher correlation.Then an improved loss function is used to correct boundary artifacts.Finally,style transfer is performed on the area to obtain the desired style transfer clothing image.(2)The iteration speed of the local clothing style transfer network is slow,and the complex network structure limits the operation speed of the algorithm;The edge information of the local target region of the generated clothing style transfer image is weakened and blurred,which affects the quality of the local clothing image style transfer.Aiming at the above two problems,a fast clothing local style transfer method based on convolutional neural network is proposed.Convert the standard convolution kernel of the residual meta-structure in the deep residual network to point-wise convolution and depth separable convolution,to drop the last 3 fully connected layers of the loss network VGG16 and the 5th layer of the convolutional layer(Block5),at the same time,the Kirsch edge detection method is used to extract the edge information of the local target area of the clothing image,and the extracted edge information is fused into the transferred clothing image.In the case of reducing network parameters,it not only improves the operating efficiency of the algorithm,but also ensures the effect of accelerating the style transfer of local clothing images after convolution.(3)The pattern,texture,material,color and other factors of the clothing are not considered in the style transfer of local clothing images,which leads to the problem of local artifacts with imprecise textures in the generated clothing images.Therefore,a method for local artifact correction of clothing images based on improved residual network is proposed.During the training process of the model,the generator is optimized by the loss function constraint and the improved ResNet50 and the discriminator,then the local style transfer image of the clothing with artifacts is input into the improved generator to obtain the local style transfer image of clothing after artifact correction.This paper hopes to perform style transfer on local clothing images through deep learning-based methods,to provide personalized clothing customization,to improve the effect of local clothing image style transfer,to improve the speed of local clothing image style transfer,and promote the leap-forward development of the clothing fashion industry.