Key Techniques For Volume Rendering Based On Programmable GPU

Data visualization technique provides intuitive and easily understandable means for the observation and the research of volume data by transforming 3D data into visible 2D image on the computer. Volume rendering attracts comprehensive attention, especially in the field of medical image processing, for it shows internal information of volume data, not just displays the surface of object. At present, researches on volume rendering mainly focus on volume rendering algorithms and interactive techniques. Volume rendering algorithms vary from one another in type of data, application fields and other factors, and none of them can be applied to all the fields. On the course of analyzing data, interactive technique can effectively mine the internal hiding information of volume date, offering protection for mastering and understanding the feature of data.After more than ten years development, volume rendering technique is far from mature and there are still some open problems to be solved in theories and applications. Based on the existing volume rendering algorithms and interactive techniques, this paper takes volume data from CT and MRI as research object, utilizes computer graphics theory and programmable technique of GPU to construct higher quality visualization algorithms and more effective interactive techniques, and combines them to provide more internal information of volume data and sufficient basis for scientific analysis of data.The main research achievements of this paper are given as follows:(1) Aiming at the hard realization of complicated illumination in existing 3D texture mapping volume rendering, vertex program and fragment program written by Cg is utilized to control the shading of fragment processor unit flexibly. The experimental results indicate that not only high quality illumination is realized, but also real-time interaction. A single-pass ray casting algorithm is proposed to reduce the complexity of casting ray parameter by GPU-based ray casting algorithm. Off-screen rendering technique is used to gain the ray casting parameter quickly by rendering volume data bounding box in this method. And GPU is utilized to realize equal-step sampling of perspective projection. Compared to the existing ones, the improved ray casting algorithm not only simplifies the render process but also flexibly realizes multiple render effects of high quality at the same time.(2) An accelerative algorithm of octree encoding method is proposed to accelerate the volume rendering algorithms. According to the depth of octree specified, volume data is fictitiously subdivided slice by slice according to the space position. In the subdivision, octree is used to store the coordinates, the maximum and minimum value of sub-blocks. The space skipping accelerating algorithm for volume rendering is realized by traversing the octree. This algorithm can also be applied to accelerate the 3D texture-based volume rendering algorithm and the ray casting volume rendering algorithm simultaneously.(3) A large-scale volume data real-time rendering algorithm based on the sub-blocks strategy is proposed to solve the storage problem caused by previous GPU volume rendering algorithms. The volume data is divided into many sub-blocks, whose size is smaller than the texture memory capacity. A same image as the one gained from direct rendering of data can be obtained by sorting sub-blocks and loading them into texture memory and rendering. With occlusion query and space skipping technology, a higher rendering speed is realized.(4) An auto-generation algorithm of transfer function based on volume histogram is given aiming at the inconveniences of existing transfer function. Firstly, the first and second derivatives of volume data are computed by GPU and their ranges are determined by 3-time standard deviation according to the central limit theory. Then, the volume histogram is constructed. Finally finds the boundary of volume data can be found based on the volume histogram and a good transfer function can be obtained. Compared to the existing algorithms, this method can automatically gains more effective transfer function without empirical value.(5) The figure stream line has to render and cut volume data twice and the gained shape can not be concave. To this end, a single-step volume clipping algorithm is developed to solve the concave object cutting problem in the two-step mode volume clipping algorithm based on the fixed graphics stream line. Firstly the frame buffer object is used to realize off-screen rendering. Then the depth information of the front and back surface of cutting body is stored into depth texture binded with frame buffer object. Finally the clipping operation is realized by judging whether the depth of volume data is in the range of depth texture or not with fragment processor. This algorithm only needs to render data once, and realizes concave clipping.(6) In order to realize better interactivity in the course of cutting operation, a cutting algorithm base on volume mark is presented. Firstly, the cutting areas are selected on the screen interactively, and the coordinate of 3D volume data are transformed from object space to screen space. Then the screen coordinate of volume data is used to judge whether the volume date belongs to the cutting area or not. Finally the cutted volume data are marked, and the cutting operation is realized by specifying the corresponding tissues as empty tissue value. A high interability without reducing the speed of rendering at the same time is obtained while the process of rendering is not changed.The research achievements above give concrete research schemes and experimental results, respectively, based on volume rendering algorithms and interactive techniques, and provide more advanced auxiliary means to analyze and understand scientific data effectively. Furthermore, the blocking strategy and automatic generation of transfer function proposed in this paper can be versatility applied to different volume rendering algorithm and enrich the basic theory and application of volume rendering.