Optimization of Visual Communication Design with Graph Convolutional Networks and Adaptive Particle Swarm Optimization

In recent days, the visual communication plays a vital role in education and information design to improve user experience and work efficiency. The traditional models usually rely on static rules and fixed layouts which struggled to adapt the complex relationships and visual contexts. Therefore, a hybrid model Graph Convolutional Networks and Adaptive Particle Swarm Optimization (GCN-APSO) is proposed for visual communication optimization design. Initially, the input data is collected from Allen Institute for Artificial Intelligence Diagrams (AI2D) Dataset. Then, GCNs are introduced to extract the high-level contextual and structural features from diagram elements which are defined as graph to understand complex relationships. Consequently, the features are then utilized to PSO-based optimization for exploring layout variations and identifies the configurations that enhance quality. Finally, the Rotation-Shift Transformer Network (RSTN) is employed to fine-tune the spatial corrections from the optimized diagrams to improve science diagram quality. From the results, the proposed GCN-APSO attained better results in terms of user satisfaction (93.74%), response time (0.28s) and operation efficiency (32.12 t/m) when compared to existing GCN-Q learning respectively.