Image style migration method based on style and content decoupling

Abstract

The invention provides an image style migration method based on style and content decoupling, and particularly relates to an image style migration neural network and a loss measurement network. The style migration neural network is any feedforward network used for the task, and the loss measurement network is composed of a style feature extraction module and a content feature extraction module. The method comprises the following training steps: sending a style migration result graph and an original content graph into a content feature extraction module of a loss measurement network, and calculating content reconstruction loss; sending the style migration result image and the original style image into a style feature extraction module of the loss measurement network, and calculating style loss; training the image style migration network through the loss measurement network, so that the generated image keeps the content of the original image and has the style characteristics of the specific style image. By the decoupling style and the content loss measurement process, the trained style migration network can effectively get rid of the interference from the content in the style graph.

Description

technical field [0001] The present invention relates to the application of deep learning in the fields of computer vision and image style transfer, in particular to an image style transfer method based on the decoupling of style and content. Background technique [0002] Image style transfer refers to the technology of using algorithms to learn the style of a specific image and rendering this style to another image without changing the main content of the picture. This technology has wide application value in image processing in social media, film and television creation aids, etc. [0003] Traditional image style transfer methods (also known as non-photorealistic rendering of images) are divided into two categories: one is computer automatic drawing to generate the required artistic style image, such as style synthesis based on color transfer and texture synthesis. The other is procedural-based simulation, which is brush-based non-photorealistic artistic rendering. The st...

AI Technical Summary

Problems solved by technology

The disadvantage of the traditional method is that the style of learning is too limited and different mathematical models need to be designed according to different styles, and the stylization effect is not satisfactory

Image neural style transfer has been widely concerned since it was proposed. Compared with traditional methods, the effect of neural style transfer is more vivid and changeable. However, the style representation in existing methods is not really suitable. The pre-trained convolutional neural network itself is used for object classification in the real environment, so that when using the pre-trained network to extract the style representation required for abstract image styles (such as: oil painting, ink painting, etc.), it is impossible to extract a suitable and beneficial style representation feature information, resulting in the final style rendering effect failing to meet expectations

Images