Visual cluster analysis is a central task to explore multidimensional data. Dimensionality Reduction (DR) techniques support this task by spatializing multidimensional (MD) data similarities as point patterns in scatterplots. However, unavoidable false and missing neighbor distortions limit their accuracy. For instance, false neighbors make truly separated data clusters appear to overlap in the layout, while missing neighbors split true clusters into falsely separated groups. In general, both types of distortions exist in DR layouts except for orthogonal linear projections (OLP) that only generate false neighbors. In this work, we propose Isometric Seriation-based Dimensionality Reductions (ISilDR) that provably generate at most missing neighbors. We study how ISilDR and OLP together could be leveraged to discover true MD clusters. An ISilDR first creates a seriation of the MD data points, i.e., an ordering along a one-dimensional projection axis, and then each pair of consecutive points along this axis is spaced by their MD distance. An $m$D ISilDR can be obtained by combining $m$ 1D ISilDRs. We study the theoretical and empirical characteristics of different variants of ISilDRs and OLPs and propose a systematic and formal analysis based on ε-neighborhood graphs. From there, we derive rules to discover cluster patterns in MD data from interactive linking of ISilDR and OLP coordinated layouts. We then conduct case studies and illustrate scenarios for trustworthy visual cluster analysis using a combination of ISilDR and other classical DR techniques.

In the early days of perceptual image quality research more than 30 years ago, the multidimensionality of distortions in perceptual space was considered important. However, research focused on scalar quality as measured by mean opinion scores. With our work, we intend to revive interest in this relevant area by presenting a first pilot dataset of annotated triplet comparisons for image quality assessment. It contains one source stimulus together with distorted versions derived from 7 distortion types at 12 levels each. Our crowdsourced and curated dataset contains roughly 50,000 responses to 7,000 triplet comparisons. We show that the multidimensional embedding of the dataset poses a challenge for many established triplet embedding algorithms. Finally, we propose a new reconstruction algorithm, dubbed logistic triplet embedding (LTE) with Tikhonov regularization. It shows promising performance. This study helps researchers to create larger datasets and better embedding techniques for multidimensional image quality. The dataset includes images and ratings and can be accessed at https://github.com/jenadeleh/multidimensionalIQA-dataset/tree/main.

A plethora of dimensionality reduction techniques have emerged over the past decades, leaving researchers and analysts with a wide variety of choices for reducing their data, all the more so given some techniques come with additional hyper-parametrization (e.g., t-SNE, UMAP, etc.). Recent studies are showing that people often use dimensionality reduction as a black-box regardless of the specific properties the method itself preserves. Hence, evaluating and comparing 2D embeddings is usually qualitatively decided, by setting embeddings side-by-side and letting human judgment decide which embedding is the best. In this work, we propose a quantitative way of evaluating embeddings, that nonetheless places human perception at the center. We run a comparative study, where we ask people to select "good" and "misleading" views between scatterplots of low-dimensional embeddings of image datasets, simulating the way people usually select embeddings. We use the study data as labels for a set of quality metrics for a supervised machine learning model whose purpose is to discover and quantify what exactly people are looking for when deciding between embeddings. With the model as a proxy for human judgments, we use it to rank embeddings on new datasets, explain why they are relevant, and quantify the degree of subjectivity when people select preferred embeddings.

We present an exploratory study on the accessibility of images in publications when viewed with color vision deficiencies (CVDs). The study is based on 1,710 images sampled from a visualization dataset (VIS30K) over five years. We simulated four CVDs on each image. First, four researchers (one with a CVD) identified existing issues and helpful aspects in a subset of the images. Based on the resulting labels, 200 crowdworkers provided 30,000 ratings on present CVD issues in the simulated images. We analyzed this data for correlations, clusters, trends, and free text comments to gain a first overview of paper figure accessibility. Overall, about 60 % of the images were rated accessible. Furthermore, our study indicates that accessibility issues are subjective and hard to detect. On a meta-level, we reflect on our study experience to point out challenges and opportunities of large-scale accessibility studies for future research directions.

A common enhancement of scatterplots represents points as small multiples, glyphs, or thumbnail images. As this encoding often results in overlaps, a general strategy is to alter the position of the data points, for instance, to a grid-like structure. Previous approaches rely on solving expensive optimization problems or on dividing the space that alter the global structure of the scatterplot. To find a good balance between efficiency and neighborhood and layout preservation, we propose Hagrid, a technique that uses space-filling curves (SFCs) to “gridify” a scatterplot without employing expensive collision detection and handling mechanisms. Using SFCs ensures that the points are plotted close to their original position, retaining approximately the same global structure. The resulting scatterplot is mapped onto a rectangular or hexagonal grid, using Hilbert and Gosper curves. We discuss and evaluate the theoretic runtime of our approach and quantitatively compare our approach to three state-of-the-art gridifying approaches, DGrid, Small multiples with gaps SMWG, and CorrelatedMultiples CMDS, in an evaluation comprising 339 scatterplots. Here, we compute several quality measures for neighborhood preservation together with an analysis of the actual runtimes. The main results show that, compared to the best other technique, Hagrid is faster by a factor of four, while achieving similar or even better quality of the gridified layout. Due to its computational efficiency, our approach also allows novel applications of gridifying approaches in interactive settings, such as removing local overlap upon hovering over a scatterplot.

We report on an interdisciplinary visual analytics project wherein automotive engineers analyze test drive videos. These videos are annotated with navigation-specific augmented reality (AR) content, and the engineers need to identify issues and evaluate the behavior of the underlying AR navigation system. With the increasing amount of video data, traditional analysis approaches can no longer be conducted in an acceptable timeframe. To address this issue, we collaboratively developed Caarvida, a visual analytics tool that helps engineers to accomplish their tasks faster and handle an increased number of videos. Caarvida combines automatic video analysis with interactive and visual user interfaces. We conducted two case studies which show that Caarvida successfully supports domain experts and speeds up their task completion time.

We propose Compadre, a tool for visual analysis for comparing distances of high-dimensional (HD) data and their low-dimensional projections. At the heart is a matrix visualization to represent the discrepancy between distance matrices, linked side-by-side with 2D scatterplot projections of the data. Using different examples and datasets, we illustrate how this approach fosters (1) evaluating dimensionality reduction techniques w.r.t. how well they project the HD data, (2) comparing them to each other side-by-side, and (3) evaluate important data features through subspace comparison. We also present a case study, in which we analyze IEEE VIS authors from 1990 to 2018, and gain new insights on the relationships between coauthors, citations, and keywords. The coauthors are projected as accurately with UMAP as with t-SNE but the projections show different insights. The structure of the citation subspace is very different from the coauthor subspace. The keyword subspace is noisy yet consistent among the three IEEE VIS sub-conferences.