GAN-based Visual Creative Generation Process Automation and Its Application Efficiency in Brand Image Design

Abstract

New social, economic, cultural, and technological trends bring new challenges to designers, and the use of data intelligence analytics is a very promising way to improve the efficiency and quality of their results. In this paper, we use convolutional neural networks to improve and optimize the GAN model, resulting in a more stable DCGAN network model. Visual innovation-related features are inputted into the discriminator in the DCGAN model, and creative visual images are automatically generated in the generator under the constraint label restriction. The network parameters of the DCGAN model are optimized through the training process to construct the brand image design process of automated visual creativity. The performance test analysis of the DCGAN model reveals that the analysis results of its IS index and FID index values are much higher than those of the WGAN and CGAN models, indicating that the generated visual creative images have high quality. The analysis of the visual creativity of the designed brand image found that the subjects’ favorability of the brand image intended by the DCGAN model reached 3.57 points on average and brought a visual feast to the subjects. This paper provides support for the application of Generative Adversarial Networks in the field of brand image design and achieves automated brand creative image design.