Research On Fast And Accurate Visualization Of Tetrahedral Volume Datasets

Direct volume rendering is an important field in scientific visualization, which helps researchers extract important scientific concepts and information from extremely large and complex volume data, and it has been widely used in many fields. Unstructured datasets, which can simulate highly detailed volumetric models with arbitrary shape and any size, are the most complicated and the effective structure in the field of scientific computing and simulation. Using unstructured datasets to represent the interior structure of a model can improve the accuracy of data and reduce the storage requirements; however, it results in the increment of complexity in computing and rendering. The dilemma of how to reduce the computing complexity and obtain high quality rendering effects has been a hot topic in the unstructured datasets volume rendering filed. Therefore, we propose some methods to visualize unstructured datasets with fast speed and high accuracy. Our main work and results of this paper are as follows:1) According to the rapid growth of large size and high resolution volumetric models, we propose a tetrahedral simplification algorithm with high quality, which optimizes error metrics, with considering the attribute variation, dihedral angles and a ratio of longest to shortest edges of the entire affected area of tetrahedral. Afterwards, we search and extract more important cells to improve the shape and the quality of targeted tetrahedra by using swap and segmentation strategy. Our method can effectively reduce the subsequent computing and rendering consumption.2) Sorting is the main bottleneck of cell projection during rendering. In order to increase the sorting effectively, we propose a K-D tree-based sorting strategy. We firstly split the entire volume data into different deep blocks, and then sort each tetrahedron in parallel within leaf nodes. Meanwhile we consider the load balancing on both sides of cutting surfaces. Our method ensures the accuracy of sorting and also significantly improves the efficiency of sorting 3) Compared to unstructured grids, regular volumetric datasets more simple and easier to handle in parallel. Converting tetrahedral volumetric datasets into regular volumetric datasets, we can improve the interactivity of the system effectively.We propose an adaptive regularization reformulation algorithm to construct the octree, and improve the sampling strategy, and then transfer the sampling results combined with the depth information into an octree texture, which can random access in GPU. Because of varying characteristics of regional depth, the sampling algorithm responds with different step-size strategy. Our method reduces the memory consumption and processing time of the data and also improves the rendering quality as well as the rendering efficiency.The experiments indicate that our methods are flexible and reliable, while also has a considerable practical value.