Research On Volume Rendering Technique For Industrial Cone Beam Computed Tomography

Volume visualization is one of the most active subfield of scientific visualization. Volume rendering is an important and effective method of volume visualization. In recent years, with the development of computer hardware and volume rendering theory, volume rendering techniques have improved tremendously. However, because of the spread of application fields of volume visualization and the improvement of application requirements, there are lots of questions to be solved in the domain of volume rendering. In different application fields, appropriate volume rendering techniques should be designed for different type of volume data, therefore, the information of structures inside of the volume data will be revealed better and the reliable evidence and guidance will be provided for analyzing the volume data scientifically. This paper takes the high resolution volume data generated from the industrial Cone Beam Computed Tomography (CBCT) as the research object, and researches the volume rendering techniques for the purpose of improving rendering quality and rendering speed combining computer graphics theory and Compute Unified Device Architecture (CUDA) technique.The main research achievements are as follows:1. As for the industrial CT volume data including severe metal artifacts, a method of transfer function design based on fuzzy classification is presented. The original information may be affected or even disappear because of metal artifacts, and traditional methods of transfer function design could not restrained the effects of metal artifacts well. By analyzing the representative form of metal artifacts, the gradient magnitude of the industrial CT volume data is modified. And then, fuzzy classification is performed combining the gray value and gradient magnitude after modification, and transfer function design is then finished. The experimental results indicate that this method could restrain the effects of metal artifacts effectively, and preserve the valuable information affected by metal artifacts, the quality of volume rendering is improved therefore.2. As for the slow rendering speed of high quality ray casting, an optimal strategy of CUDA storage models and an improved space leaping acceleration technique are presented. High quality ray casting is very slow because of larger computation amounts and higher frequency to access the texture storage. The rendering performances with different times of gradient calculation are compared firstly under the condition of different storage strategies. By analyzing the results of comparison, an optimal strategy of CUDA storage models is presented to take full advantage of GPU. In addition, the space leaping method is improved by reducing the frequency to access the texture storage for empty voxel. The experimental results illustrate that this method will significantly accelerate the rendering speed of high quality ray casting and maintenance the high quality of rendering image simultaneously.3. Aiming at the problem that the rendering speed of bricking volume rendering for large volume data will seriously descend because of switching bricking data between CPU memory and GPU memory, a model for optimal bricking is established and the optimal solution is given, and also the octree based bricking volume rendering is improved. By analyzing the Processing-Memory-Communication (PMC) properties of bricking volume rendering algorithm, a model for optimal bricking is built firstly, and optimal solution is obtained by using optimal algorithm. And then, the octree structure is used to organize the bricking data, a 3D texture named Node Code Texture is constructed to accelerate the estimation speed of validity of sampling point, and also an improved distance template is designed to leap out of invalid node efficiently. The experimental results reflect that this method could improve the rendering speed of the octree based bricking volume rendering significantly.4. As for the rapid declination of rendering speed of multi-view volume rendering for stereoscopic display of volume data, the CUDA based single-pass multi-view volume rendering method is realized and also accelerated by using both the optimal strategy of CUDA storage models and the improved space leaping acceleration technique. The rendering time of traditional multi-view volume rendering will increase along with the number of views. By firstly calculating the code of view zone of each pixel, and then setting the parameters of rays and putting the result of volume rendering onto the corresponding position of image, the result of multi-view volume rendering can be achieved by rendering only once. Combining the optimal strategy of CUDA storage models and the improved space leaping acceleration technique, the single-pass multi-view volume rendering is accelerated. The experimental results illustrate that this method could accelerate the render speed tremendously.