Title:	INTELLIGENT STYLE TRANSFER OF FILM IMAGE BASED ON CYCLEGAN
Author(s):	Da Liu, Huixin Wang and Bin Wu
ISBN:	978-989-8704-21-4
Editors:	Yingcai Xiao, Ajith P. Abraham and Jörg Roth
Year:	2020
Edition:	Single
Keywords:	High-tech Format Film, Image Style Transfer, Cycle-Consistent Generative Adversarial Networks (CycleGAN), Wasserstein GAN (WGAN), SSIM
Type:	Full
First Page:	44
Last Page:	54
Language:	English
Cover:
Full Contents:	click to dowload
Paper Abstract:	With the development and application of deep learning, image style transfer has achieved an important breakthrough. It has been able to intelligently generate stylized natural, realistic and high-quality images, which can be applied to film screens to reduce the cost of artificial effects. This paper proposes an intelligent style transfer method of film image style based on CycleGAN. In view of the high spatial resolution, large size and rich details of high-tech format film images, we have modified the input layer of the network so that the network can better process film images, and replace the original GAN loss with WGAN loss to achieve a more stable training effect. As a result, style features can be better transformed between images, and at the same time, SSIM loss is added to the cycle consistency loss to enhance the recovery of images similarity to the original image and improve the quality of the generated image. Experiments show that this method is effective in processing the style transfer of film images, and can intelligently generate high-quality and natural realistic style transfer images.
	Go Back