Data level comparisons of direct volume rendering algorithms

Identifying and visualizing uncertainty together with the data is a well recognized problem. One of the culprits that introduce uncertainty in the visualization pipeline is the visualization algorithm itself. Uncertainties introduced in this way usually arise from approximations and manifest themselves as artifacts in the resulting images. In this work, we focus on comparing different direct volume rendering (DVR) algorithms and their artifacts as a result of DVR algorithm selections and their associated parameter settings. We present a new data level comparison methodology that uses differences in intermediate rendering information. In image level comparisons, quantized pixel values are the starting point for comparison measurements. In contrast, data level comparison techniques have the advantage of accessing and evaluating the intermediate 3D information during the rendering process. Our data level approach overcomes limitations of image level approaches and provides capabilities to compare application dependent details as well as general rendering qualities. Our methodology consists of 4 main components: a list of algorithm specification item and rendering parameters, standard test data sets, a base algorithm approach, and data level metrics and their visual mappings.; One of difficulties in comparisons stems from the numerous rendering parameters and algorithm specifications involved. We discuss the list of algorithm specification items one should look for in comparison studies of DVR algorithms. We also present a checkerboard and a ramp test data specification as a standard for comparing and differentiating algorithms. One of the key challenges with our data level comparison approach is finding a common base for comparing the rich variety of DVR algorithms. Therefore, we propose to use a base algorithm approach in which the given algorithm is mapped to a base algorithm with the given algorithm specifications to be compared at the data level. In addition, the data level metrics are derived to quantify the comparison measurements among DVR algorithms. We also present a couple of visual mappings (or visualization methods) to be used for these metrics. We demonstrate the superiority of our data level approach over the conventional image level approach by using a number of examples.