An Image-based Typology for Visualization

TL;DR

This paper addresses the lack of appearance-based categorization for visualization images by introducing VisTypes, an image-based typology derived from a qualitative analysis of 6,833 figures from IEEE VIS papers. The authors define visual stimuli and develop a 6-phase coding process, resulting in 10 distinct visualization types, two image purposes, and 2D/3D dimensionality labels, validated via a controlled expert study showing high inter-rater agreement. The typology enables robust analysis of design styles, historical evolution, and standardization efforts, and it provides a dataset and tools to explore visual representations across time and context. Practically, VisTypes supports broader evaluation, cross-disciplinary research, and potential standardization in visualization practice and education.

Abstract

We present and discuss the results of a qualitative analysis of visualization images to derive an image-based typology of visualizations. For each image, we seek to identify its main focus or the essential stimuli. As a result, we derived 10 image-based visualization types. We describe coding decisions we made in the derivation process. The resulting image typology can serve a number of purposes: enabling researchers and practitioners to identify visual design styles, facilitating the categorization of visualization images for the purpose of research and teaching, enabling researchers to study the evolution of the community and its research output over time, and facilitating a discussion of standardization in visualization. In addition, the tool and dataset enable scholars to closely examine the images and how they are published and communicated in our community. osf.io/dxjwt presents a pre-registration and all supplemental materials.

Figures (26)

• Figure 1: Our 10 visualization types (VisTypes). They characterize the associations of what we see from images. Here we show the overall image counts and relative frequency trends in % for 1990--2020. Surface-based techniques & volumes are most common, followed by line-based techniques and points. Together, these three main types make up about half of the images we coded (See our online tool https://visimagenavigator.github.io/ for additional examples). Each type represents a family of visual appearances of essential stimuli, the main focus of a visualization image.
• Figure 2: Code type comparisons. We compare codes between Borkin et al. borkin2013makes's types color-coded in cyan and ours (color-coded in orange) in (a)-(c) and between Deng et al. deng2020visimages's types (color-coded in green) and ours in (d). We label image (a) afzal2012spatialtext---because the essential stimuli that represent the geospatial data and metadata here are text. Our typology does not include "map" because maps are a representation technique that may have very different visual appearances (e. g., consist of only points, areas, or text as seen here). We label image (b) byron2008stacked as a schematic representation, elucidating its conceptual illustration; actual data are not the main focus. Image (c) has two labels Bar and Point. We have more complete essential stimuli to signify the importance of seeing both Bar and Point. Image (d) yue2018bitextract emphasizes data types (networks), and we use Bar & Node-link, beyond the technique label of Donut-chart.
• Figure 3: The image coding process. We developed the method, identified the type codes, and performed the image labeling and curation in a multi-year team effort. This lengthy process includes three failed attempts (Phases 1 $\&$ 2). We finally used, as shown in Phases 3--6, a similarity-based appearance categorization to derive the essential stimuli. Observation: visualization techniques and design elements are distinct from what we see for making conscious judgments of representations.
• Figure 4: The proportion of applied image codes for each category, relative to the total 4,070 visualization images (after excluding the pure schematic and GUI images). We can see that the most common visualization types were "generalized sur-face and volume representations" and "line-based representations."
• Figure 5: Percentage of 2D3D by total images. The sum in each year is larger than $100\%$ because some images contain both types.