Interactive Rendering Of Large Volume Data

Direct Volume Rendering (DVR) is an important method for volume data visualization. Incontrast to other rendering techniques, DVR is able to represent the inner structures of thevolume, and the result has higher quality and lower distortion. In addition, DVR is able toproduce the rendered image directly without reconstructing the volume with proxy geometricprimitives. However, the cache capacity and bandwidth limitation of the computer GraphicProcessing Unit, lead to the difficulty of interactive rendering of large volume data withtraditional volume rendering techniques. Therefore, the interactive rendering of large volumedata has become a hot area of scientific visualization research nowadays.In order to effectively reduce the amount of data to complete the interactive rendering oflarge scale data, we systematically analyze and process the volume data in the preprocessingstage. First, volume data is subdivided into blocks with equal size; then we adopt adaptivebricking level of detail(LOD) technique to obtain the data importance of block by analyzing thevariance weighted Shannon entropy, and compute the dynamic threshold as the LOD of blockrespectively; we then achieve the compressed multi-representation texture as codebook for thevolume data by applying a merge operation to the bricking texture, and load the texture intographics card cache for sampling; then we introduce a multi-resolution block texture searchmode to find the specific sampling points in the codebook, and adopt the inter-blockinterpolation to sample the adjacent blocks with different LOD, which ensure the correctsampling for the boundary regions of blocks; finally, we adopt ray casting based on GPU torender the final image.In this paper, an adaptive bricking LOD selection algorithm is proposed to obtain thedynamic threshold according to the characteristics of the block scalar data. The LODdetermination is more reasonable. Thus our approach avoid the case that all of the blocks have asimilar threshold to determine the LOD, which the volume data could not be compressed andretained the detail of blocks effectively. And the method of multi-resolution block texture search reduces the swapping cost for texture data between main memory and texture memory, thus itimproves interactive frame rate of rendering.In the experiment, this paper do not choose the ultra-large-scale volume data due to thelimitation of the experimental hardware but better analyzing the performance of algorithm usingfour representative volume data to validate the experiment which show the merits of thealgorithm sufficiently instead. Our experiments demonstrate the higher quality and higherrendering efficiency of the proposed methods.