GPU-based Ray-casting Volume Rendering

Visualization in scientific computing(VISC) is widely used in medicine, geological exploration, meteorology, et al. In the study methods of VISC, ray-casting plays an important role because of the high-quality reconstruction images. However, the traditional CPU-based ray-casting is too slow for wide use. Therefore, research on how to improve the rendering speed and quality of ray-casting is significant to perfect visual theory and expand its application areas.Aiming at the disadvantage of the multi-pass ray-casting, a single-pass ray-casting algorithm is implemented firstly in this thesis. Rendering front face and back face once can get the ray parameters. Only one fragment program is needed to generate entry point and ray direction, implement ray marching, ray termination, color and opacity accumulation, et al. Experimental results show the algorithm can reduce the complexity of ray casting effectively.In order to improve the rendering speed, an acceleration algorithm is proposed based on space leaping technology in this thesis. Octree structure is used to organize volume data. After obtaining the data statistics, the voxels non-contributing to final image can be skipped according to the current transfer function, then the rendering is speeded up. Experimental results show that the algorithm can effectively skip the internal non-contributing voxels, and significantly speed up the rendering of volume data which contains a large number of non-contributing voxels internal.The acceleration algorithm based on space leaping can't speed up the rendering of the volume data which contains only a small number of non-contributing voxels internal, and two additional textures are needed, which takes up memory resources. Therefore another acceleration algorithm is proposed based on compact bounding box, which derives from the thinking of volume data subdivision and the point of keeping load balance of vertex processor and fragment processor. Volume data is subdivided with octree structure, and compact bounding box is generated according to the current transfer function. So the voxels non-contributing to final image are excluded. Experimental results illustrate the effectiveness of the algorithm.On the basis of basic algorithm and acceleration algorithm several advanced algorithms are implemented, which include maximum intensity projection, discrete iso-surface reconstruction and curvature reconstruction based on cubic B-spline filter.