Shuning Jiang¹, Meng Ling², Rober S. Laramee³, Michael Sedlmair⁴, Jian Chen¹; ¹The Ohio State University, USA, ²The Washinton Post, USA, ³University of Nottingham, UK, ⁴University of Stuttgart, Germany

Words people use have long served as a window into perception. Despite decades of research on visual complexity (Oliva et al. 2004; Rosenholtz et al. 2005; Chu et al. 2025), most quantitative metrics are derived from natural scenes rather than data visualizations. To fill in the gap, we focus on the degree to which specific features, attributes, and factors are more or less relevant. We discuss lay terms, provide definitions, and suggest their potential effects on observers, by conducting a qualitative analysis of ~700 human verbal descriptions of visualization image pairs to identify the perceptual dimensions that observers spontaneously use when describing complexity. From these descriptions, supported by human expert coding and LLM-assisted analysis (GPT-5mini), we developed and defined a lexicon of six dimensions: (1) Data Density and Clutter, (2) Visual Encoding Clarity, (3) Color, Symbol, and Texture Detail, (4) Schema, (5) Semantics and Text Legibility, and (6) Aesthetics. We mapped these dimensions to expressions of Immediacy and Cognitive Load (e.g., “unclear where to look,” “takes longer to interpret”) and linked them to memorability outcomes, following that of Kyle-Davidson et al. 2024. Using this lexicon, we characterized the complexity of ~700 visualization images and provided the annotated dataset. In contrast to findings from natural scenes we observed no strong relationship between complexity and memorability in visualization images (r-squared=0.07). Our results establish a principled vocabulary for perceptual drivers of visualization complexity and offer a foundation for developing objective, computational measures tailored to data visualization images.

Acknowledgements: Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) under Germany’s Excellence Strategy – EXC 2120/1 – 390831618; DFG, German Research Foundation) – Project-ID 251654672; EPSRC UKRI157, APP17227