Comic Image Style Transfer Based on De-GAN
Zuoyun Yang, Hongqiong Huang · Atlantis Highlights in Computer Sciences/Atlantis highlights in computer sciences · 2023
There are still many problems in the current comic style transfer method, such as the style of the generated image does not conform to people's aesthetics, the color is far from the original image, and so on.This paper proposes a new network architecture based on the idea of generative adversarial networks.For the generator, the Desnet module is introduced in the feature conversion layer, which reduces the amount of network parameters while optimizing the efficiency of feature extraction.For the discriminator, this paper introduces layer normalization to denoise the image to solve the problem of image artifacts.In terms of loss function, this paper introduces the color reconstruction loss item to supplement the original loss function, which improves the color of the generated comic image and makes it closer to the original painting.The experimental results show that compared with the current mainstream generative adversarial network, the network model in this paper has achieved better results in the field of comic style transfer.