A convolutional neural networks based approach for clustering of emotional elements in art design

Abstract

The rapid advancement of industrialization has sparked the emergence of diverse art and design theories. As a trailblazer in the realm of industrial art and design theory, visual communication has transcended the boundaries of merely arranging and combining individual elements. Embracing the potential of artificial intelligence technology, the extraction of multidimensional abstract data and the acceleration of the art design process have gained considerable momentum. This study delves into the abstract emotional facets within the methodology of visual communication art design. Initially, convolutional neural networks (CNN) are employed to extract expressive features from the poster's visual information. Subsequently, these features are utilized to cluster emotional elements using a variational autoencoder (VAE). Through this clustering process, the poster images are categorized into positive, negative, and neutral classes. Experimental results demonstrate a silhouette coefficient surpassing 0.7, while the system framework exhibits clustering accuracy and efficiency exceeding 80% in single sentiment class testing. These outcomes underscore the efficacy of the proposed CNN-VAE-based clustering framework in analyzing the dynamic content of design elements. This framework presents a novel approach for future art design within the realm of visual communication.