Arbitrary style migration method and system, storage medium, computer equipment and terminal

Abstract

The invention belongs to the technical field of generative adversarial networks and style migration in computer vision, and discloses an arbitrary style migration method and system, a storage medium, computer equipment and a terminal.The input of an arbitrary style migration system network is composed of a content image Ic and a style image Is, and the output of the arbitrary style migration system network is a stylized image Ics; the network structure is composed of an image encoder Enc, an adaptive channel network AdaNet, a decoder Dec, a discriminator D and an auxiliary classifier Q, the arbitrary style migration method comprises the following steps: training an arbitrary style migration network model by utilizing a training data set; and after the training of the network model is completed, network feedforward is carried out by using an image encoder Enc, an adaptive channel network AdaNet and a decoder Dec, so that a stylized image can be obtained. According to the method, the visual performance of the generated artistic image is improved, the defects of fuzzy blocks, color blocks, meaningless lines and the like existing in the image generated by the existing method are eliminated, and clearer image details are achieved.

Description

technical field [0001] The invention belongs to the technical field of generative confrontation network and style transfer in computer vision, and in particular relates to an arbitrary style transfer method, system, storage medium, computer equipment and terminal. Background technique [0002] At present, style transfer is a technique that uses algorithms to learn the style of artistic images, which can transform an image into another artistic style while maintaining the invariance of the local structure of the content image. In 2016, Gatys et al. creatively proposed the use of deep convolutional networks to complete the conversion from real images to artistic images, and called this technique "style transfer". Style transfer technology has been widely used in photography, camera software and other applications. It is foreseeable that with the further development and improvement of this technology, it will have broad application prospects in industrial scenes such as film sp...

AI Technical Summary

Problems solved by technology

[0007] (1) The existing arbitrary style transfer algorithm cannot decouple the content information and style information in the image, and it is difficult to balance the content structure and stylization characteristics in the generated artistic image

[0008] (2) There are a lot of defects such as blurred blocks, color blocks and meaningless lines in the artistic image generated by the prior art method, and the conversion result at high resolution is not good

[0009] (3) The generalization ability of the prior art method on the artistic style category outside the training set is poor, the stylization characteristics are not prominent and the content details are lost more

[0012] (2) It is difficult to find the direct cause of the visual defects in the artistic images generated by the style transfer algorithm, and the existing algorithms do not perform well at high resolutions, so eliminating image visual defects and improving the algorithm’s generation results at high resolutions can The most intuitive way to improve visual performance

Images