Human face cartoon image generation based on CycleGAN
Xu Xiao, Xianlan Wang · 5th International Conference on Computer Information Science and Application Technology (CISAT 2022) · 2022
In recent years, the rapid development of deep learning has greatly promoted the progress of computer vision and image graphics. Many new image style transfer algorithms based on deep learning have made a huge leap in image generation quality and operation speed. Online animation is the expression of animation and comics, including animation books, films, television, audio-visual products, stage plays and the development of animation products based on modern information and communication technology. However, most of the cartoon faces are similar, and the facial features are single, but if a more detailed description is carried out, the production cost is high. In this paper, a cycle-consistent generative adversarial network is used to convert the real face image to the comic style face image. The network does not require paired matching data, so that the network can use a wide range of data and is less difficult to obtain. Through experiments, it is found that the improved CycleGAN model has a good effect on generating cartoon face images, which can save a lot of cost for cartoon makers, and the generated cartoon faces have the advantages of being more realistic and detailed.