Details

Autor(en) / Beteiligte

• Zhao, Wenqing
• Zhu, Jianlin
• Li, Ping
• Huang, Jin
• Tang, Junwei

Titel

Attention mechanism-based generative adversarial networks for image cartoonization

Ist Teil von

• The Visual computer, 2024-06, Vol.40 (6), p.3971-3984

Ort / Verlag

Berlin/Heidelberg: Springer Berlin Heidelberg

Erscheinungsjahr

2024

Link zum Volltext

Quelle

Alma/SFX Local Collection

Beschreibungen/Notizen

• As a common art form in daily life, cartoon images play an important role in the fields of movie production and science education. However, intelligently generating different style cartoon images from real-world photographs often has a number of problems, which mainly include: (1) The generated images do not have obvious cartoon-style textures; (2) the generated images are prone to structural confusion, color artifacts, and loss of the original image content. Therefore, style transfer and preservation of original content is a great challenge in the field of image cartoonization. In this paper, we propose an attention mechanism-based generating adversarial networks for image cartoonization to address the above problem. The method uses the attention module to perform feature correction on the deep network features extracted from the residual blocks in the generative model, so that it strengthens the cartoon features of the generated image and enhances the ability of the generative model to perceive the cartoon style. At the same time, we also use the attention module to perform feature correction on the features extracted from the convolutional block in the discriminative model, so that it can strengthen the discriminative ability between the generated cartoon image and the real cartoon image while reducing the structural confusion, color artifacts, and loss of the original image content caused by style transfer. Qualitative experiments and quantitative evaluations demonstrate the advantages of our method in terms of style transfer and content preservation, while ablation study validates the role of each module in the method. The code is https://github.com/zwq11/Image-Cartoonization.git .