GPU Based High Quality Interactive Volume Visualization

The techniques for volume visualization have been one of the active topics over the last decade. With rapid progress of modern medicine, larger and larger scale volumetric datasets demand on more effective and efficient visualization algorithms. Recently, surface extraction techniques are gradually taken placed by direct volume rendering, which can visualize the volumetric information with a semitransparent realistic effect. There are two challenges on direct volume rendering. Firstly, due to the high computational expense of accurate ray integration, high quality image and optimized performance can not be obtained simultaneously. Secondly, the complexity of the volumetric dataset often increases the difficulty on the efficient volume analysis. In this thesis, we focus on the techniques of GPU (Graphics Processing Unit) based volume rendering. We propose to seek for a tradeoff between the realistic effects and interactive performance. The main contributions of this thesis are listed in the following four respects:1. Volume Shading: We propose a novel algorithm for texture-based volume rendering based on the per-pixel shading which can improve the rendering quality greatly. The pixel-wise accurate shading is produced for the arbitrary and dynamically changing directional light source using the normalized gradient during the lighting. Furthermore, the flexible GPU functionality is utilized to enable fast data loading and interactive rendering.2. Transfer Function Design: We propose a mechanism to induce interactive spatial multi-dimensional transfer functions, which can assign color and opacity to the volume based not only on the scalar intensity of the dataset but also on the position of voxels. The proposed technique is implemented by the GPU texture functionality to enable interactively select and explore regions of the volumetric dataset.3. Data-Centered Volume Analysis and Rendering: To distinguish the data structures efficiently, we propose a multi-object hybrid volume rendering algorithm based on image segmentation. The effect volume analysis is archived by data classification. The proposed technique is implemented on the GPU pixel processor to provide the reasonable performance. Moreover, an efficient interactive volume clipping method based on the dataset region marking is proposed to expose the hidden important information. The proposed technique is integrated into the graphics hardware to accelerate volume rendering.4. Rendering Pipeline: We propose a modified pipeline to load the dataset dynamically, and reside only one data brick on texture memory during the rendering. The method is very efficient for rendering large-scale volumetric dataset interactively on general purpose PCs. Using the GPU as a co-processor, the gradient is estimated on the fly to avoid the memory consumption.