Generative Adversarial Network-Based Experience Design for Visual Communication: An Innovative Exploration in Digital Media Arts

Abstract

In the course of technological advancement, the landscape of artistic media creation has witnessed a proliferation of diversity. The integration of artificial intelligence has infused art media with pioneering elements, giving rise to emerging domains like generative art and algorithmic art, thereby infusing artistic expression with a forward-looking essence. This study delves into the realm of cross-modal artistry within digital media art creation, introducing a TAE-GAN image generation framework built upon GAN architecture. The framework initiates the process by employing TextCNN with self-attention to extract emotional features and embed word vectors from textual data. Subsequently, utilizing the generated feature vectors, the model’s output is realized through the GAN network, facilitating image generation based on textual data. Finally, quantitative metrics such as IS and FID are employed for model evaluation, along with user satisfaction analysis regarding text-image synthesis. Experimental findings underscore the commendable performance of the framework across both publicly available and custom datasets, attributed to its multifaceted network structures. Particularly noteworthy is the satisfaction rate for text-image alignment within the self-built dataset, exceeding 60%, presenting a pioneering methodological framework for future digital media art creation.

Keywords

• Emotion analysis
• GAN
• digital media
• deep learning
• TextCNN