For 3D scatterplots, we present an interpolation and projection technique that supports the smooth exchange of one or two data dimensions at a time. Even though this exchange can be considered as a rotation in 4D or 5D data domains, we guarantee that the projection to image space is perceived as a 3D rigid body rotation--with a consistent motion of the data points. We conducted a controlled user study showing that 3D rigid body rotations outperform direct transition between scatterplots. We further extend our technique to support navigation between 3D scatterplots by introducing 3D scatterplot matrices. The usefulness of our approach is demonstrated by application examples, including a case study with a natural language processing expert.

Abstract Anaglyph stereo provides a low‐budget solution to viewing stereoscopic images. However, it may suffer from ghosting and bad color reproduction. Here we address the first issue. We present a novel technique to perceptually calibrate an anaglyph stereoscopic system and to use the calibration to eliminate ghosting from the anaglyph image. We build a model based on luminance perception by the left and right eyes through the anaglyph glasses. We do not rely on power spectra of a monitor or on transmission spectra of anaglyph glasses, but show how the five parameters of our model can be captured with just a few measurements within a minute. We present how full color, half color, and gray anaglyphs can be rendered with our technique and compare them to the traditional method.

Abstract In contrast to 2D scatterplots, the existing 3D variants have the advantage of showing one additional data dimension, but suffer from inadequate spatial and shape perception and therefore are not well suited to display structures of the underlying data. We improve shape perception by applying a new illumination technique to the pointcloud representation of 3D scatterplots. Points are classified as locally linear, planar, and volumetric structures—according to the eigenvalues of the inverse distance‐weighted covariance matrix at each data element. Based on this classification, different lighting models are applied: codimension‐2 illumination, surface illumination, and emissive volumetric illumination. Our technique lends itself to efficient GPU point rendering and can be combined with existing methods like semi‐transparent rendering, halos, and depth or attribute based color coding. The user can interactively navigate in the dataset and manipulate the classification and other visualization parameters. We demonstrate our visualization technique by showing examples of multi‐dimensional data and of generic pointcloud data.