Article: Adversarial learning-based image generation algorithm for AI art creation Journal: International Journal of Information and Communication Technology (IJICT) 2025 Vol.26 No.4 pp.57 - 71 Abstract: Generative adversarial network (GAN) is widely used in AI art creation image generation tasks. Focusing on the issue of poor quality of images generated by existing algorithms, this paper firstly optimises the GAN by minimising mutual information (MIGAN), captures the features of the image by adopting multiscale null convolution, and employs the spatial feature transformation to ensure the completeness of semantic information of the image. Then self-attention and channel attention are introduced in MIGAN to focus on the important features of the image from both spatial and channel dimensions to finally get the generated image. Adversarial learning between the generated image and the real image through the discriminator, and designing the adversarial loss function to optimise the network model, which effectively improves the image generation effect. Finally, the algorithm is experimentally verified to produce high-quality AI art creation images with at least 33% reduction. Inderscience Publishers - linking academia, business and industry through research

Title: Adversarial learning-based image generation algorithm for AI art creation

Authors: Zhou Chen; Zaixi Xia

Addresses: Chengdu Academy of Fine Arts, Sichuan Conservatory of Music, Chengdu, 610021, China ' Digital Media Arts College, Chengdu Vocational University of the Arts, Chengdu, 611433, China

Abstract: Generative adversarial network (GAN) is widely used in AI art creation image generation tasks. Focusing on the issue of poor quality of images generated by existing algorithms, this paper firstly optimises the GAN by minimising mutual information (MIGAN), captures the features of the image by adopting multiscale null convolution, and employs the spatial feature transformation to ensure the completeness of semantic information of the image. Then self-attention and channel attention are introduced in MIGAN to focus on the important features of the image from both spatial and channel dimensions to finally get the generated image. Adversarial learning between the generated image and the real image through the discriminator, and designing the adversarial loss function to optimise the network model, which effectively improves the image generation effect. Finally, the algorithm is experimentally verified to produce high-quality AI art creation images with at least 33% reduction.

Keywords: image generation; generative adversarial network; GAN; multiscale null convolution; self-attention mechanism; channel attention.

DOI: 10.1504/IJICT.2025.144655

International Journal of Information and Communication Technology, 2025 Vol.26 No.4, pp.57 - 71

Received: 08 Dec 2024
Accepted: 18 Dec 2024
Published online: 25 Feb 2025 *

Title: Adversarial learning-based image generation algorithm for AI art creation

Keep up-to-date