What Makes a Visualization Image Complex?

Area of Science:

• Data Visualization
• Human-Computer Interaction
• Perceptual Science

Background:

• Perceived visual complexity (VC) is crucial for effective data visualization.
• Objective metrics are needed to quantify VC and understand user perception.
• Existing metrics may not fully capture the nuances of VC in diverse visualizations.

Purpose of the Study:

• To investigate perceived visual complexity in data visualizations using objective image-based metrics.
• To evaluate the alignment of human-scored VC with various computational metrics.
• To develop and validate a quantifiable model for predicting perceived VC.

Main Methods:

• Conducted a large-scale crowdsourcing experiment with 349 participants rating 1,800 visualization images for VC.
• Analyzed 12 image-based metrics, including pixel-based, clutter, color, shape, and novel object-based metrics (meaningful-color-count (MeC) and text-to-ink ratio (TiR)).
• Developed a quantification model grounded in the VisComplexity2K dataset.

Main Results:

• Both low-level features (edges, corners) and high-level elements (distinct colors) significantly influence perceived VC.
• Feature congestion is a strong predictor for visualizations with continuous color/texture, while edge density is effective for node-link diagrams.
• A bell-curve relationship was observed for text-to-ink ratio (TiR), indicating an optimal text density for reducing complexity.

Conclusions:

• Objective metrics, including novel ones like MeC and TiR, can effectively predict perceived visual complexity in data visualizations.
• Understanding the interplay of low-level and high-level features provides insights into visualization design.
• The developed model offers interpretable, metric-based explanations for perceived VC, bridging computational approaches and human perception.