Projection-based Data Level Comparisons of Direct Volume Rendering Algorithms

abstract 0 The focus of this paper is on comparative visualization tools for scientists and also on a mechanism for communicating their results on the web. Direct volume rendering (DVR) is one of the most popular methods for visualizing 3D scientiic data including medical and molecular volumetric data. Since DVR is a relatively expensive method, much research has focused on making it faster as well as making it work with various grid structures. Unfortunately, this has lead to a host of DVR algorithms that produce images that are slightly diierent from each other. In situations where the diierences in the resulting DVR images makes a diierence { for example, in medical imaging where slight diierence may lead to a mis-diagnosis, or in molecular studies where it is important to understand the binding strengths, visualization users must understand the limitations of each DVR algorithm. Previous eeort in highlighting these diierences have focused on image level comparisons. This paper advocates for data level comparison. That is, we take advantage of intermediate 3D information available during the rendering process to do the comparison. The main beneet of this approach is that it not only allows us to identify the regions and extents of diierences among DVR algorithms, it also provides us a mechanism for explaining why they are diierent. One of the main challenges and contribution of this work is in nding common bases for comparing diierent DVR algorithms. In this paper, we describe the projection based algorithms as the basis for comparison. Using this basis, diierent comparison metrics are derived and used to evaluate diierent DVR algorithms. We illustrate the eeectiveness of this approach using several case studies. The second focus of this paper is on communicating these results on the web. We provide a web interface for users to select diierent DVR algorithms and parameters for comparison, as well as diierent view angles and a set of volumetric data sets. Since the results of data level comparisons may be images or 3D analyses, the comparisons are sent as either GIF images and/or VRML les.