Study On Efficient Visualization Of Tetrhedral Volume Datasets

Volume visualization can represent the significant interior structure or information of data field in three dimensions. It has been implemented in medical image analysis, geological data vi-sualization, computational fluid dynamics simulation, and so on. General volume data structure can be divided into two categories:regular datasets and unstructured datasets. Regular datasets often has been constructed by an array in three dimensions, sampling on grid points which have same span with each other. Regular datasets can be shown simply and handled easily, being ap-propriate to uniform data field. But most physical simulations, for example computational fluid dynamics simulation or partial differential equations, generate data field which is nonuniform. Us-ing regular datasets will bring about accuracy problem and redundant storage. So unstructured datasets have been represented by polyhedron or other unstructured data elements, and tetrahedra has been used popularly. Unstructured volume representation of data field in three dimensions im-proves accuracy and reduces requirement in storage, but it also increases the cost in computation and rendering. How to improve the rendering efficiency and enhance the rendering accuracy, are two primary challenges in rendering unstructured datasets. This paper made a lot of improvements as following:Regular datasets can reduce a lot of computation cost because its structure is simple and easy to handle in parallel. It also does not need intersection computation. Therefore, regu-larization is the most direct scheme to solve this problem. Some works transform tetrahedral datasets into octree structure and render it based on texture. If the unstructured datasets have higher accuracy, regularization will generate the octree with more layers which increase the time to store and find. We propose a dual-structure scheme to decompose the datasets in-to two components, which can be represented as two GPU-friendly textures. The new data structure is memory-efficient, GPU-friendly, and thus allows for effective GPU-based vol-ume visualization. Regularization is a method with accuracy loss. The main work of this paper is still about rendering directly on unstructured datasets. Projection is a typical method. Its bottleneck of rendering is sorting. Sorting value of tetrahedral centroids transform geometry sorting into float value sorting, which is easy and convenient. Until now, this kind of sorting has been handled by general sorting method. But our method considers the spatial coherence of tetrahedral datasets. We pre-compute the distribution of tetrahedral datasets in different deep blocks under different views. Tetrahedral datasets have been divided into corresponding block quickly with load balancing. The efficiency of sorting centroids increases with our method.Sorting centroids is not accurate. It will produce some rendering error because of wrong order. Sorting on exact occlusion relationship can solve this problem, but it is a great chal-lenge to parallel. We solve this problem by layering the datasets, but the order of the layers is still processed seriously. The order between layers can be broken by physical partition. The tetrahedral datasets have been organized into a spatial octree structure. The tetrahedra on the partition plane has been divided into small ones. The tetrahedra in different node has been sorted independently, that increases the sorting efficiency. The physical partition needs more calculation and the more tetrahedra after partition produces many difficulties in storage and rendering. We improves this method and proposes a scheme named logical partition, which divides the tetrahedral datasets into different deep block without real partition. Only corresponding area has been integrated when rendering. This method increases the efficiency without more storage requirement.The quality of rendering unstructured datasets is another important challenge. Most of ren-dering systems on unstructured datasets suppose that the interpolation in one element is linear, whose accuracy is limited. This paper reconstructs datasets with7direction box s-pline and builds a data field with high-order continuity. Then quadratic tetrahedra has been extracted and rendered by finite element methods. The experiments show that rendering with quadratic interpolation can increase the accuracy of result image, enhance the features and reduce the crack between tetrahedra. The quality of result image has been improved a lot.The contributions of this paper are on visualization of unstructured datasets. The content has been divided two categories:one is transforming unstructured datasets into regular datasets; the other is rendering directly on unstructured datasets. In the former part, this paper proposes a dual-structure scheme which can balance between storage and accuracy. In the later part, this paper proposes a complete frame based on tetrahedra projection. The sort and the rendering have been separated. The efficiency of sorting tetrahedral centroids has been improved by using spatial coherence. The time of accurate sorting has been reduced by physical partition and logical parti-tion. In rendering aspect, this paper reconstructs the linear data field and improves the accuracy by using quadratic interpolation. This frame is complete, improving both rendering efficiency and quality. It brings great help to identifying and extracting feature in scientific computation. A lot of experiments identify the reliability of our system.