Video image optimization-based style transformation method

Abstract

The invention provides a video image optimization-based style transformation method. The method mainly comprises style transformation in a still image, optimization-based coherent video style transfer and training. According to the process of the method, random Gaussian noises are adopted to perform initialization, a content loss function and a style loss function are defined, and therefore, the energy minimization problem of the video style transfer method can be solved, long-term consistency and image quality during the movement of a camera can be improved; and a network with parameters is adopted as input frames, previously generated frames are distorted and masked by blocking masks, output is generated, and the function is adopted to obtain a video recursively. With the method of the invention adopted, the loss functions of the short-term consistency and long-term consistency of the stylized video and the multi-channel mode of the stylized video can be realized, namely, and a stable result is generated under fast motion and intensive blocking conditions; operating time is greatly reduced; and the quality of images can be improved.

Description

technical field [0001] The invention relates to the field of image and video style conversion, in particular to a style conversion method based on video image optimization. Background technique [0002] Video image processing is one of the hot topics in the field of computer vision research. This year, a new technology based on deep learning-video image style conversion has attracted more and more attention. It uses computer as a tool to simulate the drawing styles of different art forms with algorithms, and enhances the expression of visual information in video images. This technology that effectively combines computer technology and artistic aesthetics is more and more popular among users. It has a wide range of applications, such as providing inspiration or creative help for artists, image editing of various artistic styles on mobile devices, stylization of video images in entertainment industries such as movies, games, and animation, as well as in the fields of scientifi...

AI Technical Summary

Problems solved by technology

[0004] For problems such as long running time, the purpose of the present invention is to provide a style transfer method based on video image optimization, use random Gaussian noise for initialization, define content loss function and style loss function, and solve the energy minimization of video style transfer method Problem, improve long-term consistency and image quality during camera motion, use a network with parameters as input frames, occlusion mask to distort and mask previously produced frames and produce output, apply this function recursively to get a video

Images