Mark A. Livingston

Multivariate visualization techniques have attracted great interest as the dimensionality of data sets grows. One premise of such techniques is that simultaneous visual representation of multiple variables will enable the data analyst to detect patterns amongst multiple variables. Such insights could lead to development of new techniques for rigorous (numerical) analysis of complex relationships hidden within the data. Two natural questions arise from this premise: Which multivariate visualization techniques are the most effective for high-dimensional data sets? How does the analysis task change this utility ranking? We present a user study with a new task to answer the first question. We provide some insights to the second question based on the results of our study and results available in the literature. Our task led to significant differences in error, response time, and subjective workload ratings amongst four visualization techniques. We implemented three integrated techniques (Data-driven Spots, Oriented Slivers, and Attribute Blocks), as well as a baseline case of separate grayscale images. The baseline case fared poorly on all three measures, whereas Datadriven Spots yielded the best accuracy and was among the best in response time. These results differ from comparisons of similar techniques with other tasks, and we review all the techniques, tasks, and results (from our work and previous work) to understand the reasons for this discrepancy.

Multi-valued data sets are increasingly common, with the number of dimensions growing. A number of multi-variate visualization techniques have been presented to display such data. However, evaluating the utility of such techniques for general data sets remains difficult. Thus most techniques are studied on only one data set. Another criticism that could be levied against previous evaluations of multi-variate visualizations is that the task doesn't require the presence of multiple variables. At the same time, the taxonomy of tasks that users may perform visually is extensive. We designed a task, trend localization, that required comparison of multiple data values in a multi-variate visualization. We then conducted a user study with this task, evaluating five multivariate visualization techniques from the literature (Brush Strokes, Data-Driven Spots, Oriented Slivers, Color Blending, Dimensional Stacking) and juxtaposed grayscale maps. We report the results and discuss the implications for both the techniques and the task.

Low-visibility conditions for navigation of vehicles are a frequent occurrence. Driving at night, in blizzards, in sand storms, or in fog form an obvious set of challenging conditions. Remote operation of unmanned vehicles through a camera image provides a similar difficulty. Gathering intelligence from satellite imagery can similarly benefit from improved visibility. Advanced image processing techniques (e.g., contrast enhancement or tone mapping) purport to improve the perceptual quality of images that lack the contrast or color depth perceived by the human visual system (HVS). Applying such an algorithm intelligently to these low-visibility conditions gives us the ability to provide a perceptually usable assisted-vision system.