電機資訊學院: 資訊網路與多媒體研究所指導教授: 莊永裕蘇彥禎Su, Yan-JenYan-JenSu2017-03-062018-07-052017-03-062018-07-052015http://ntur.lib.ntu.edu.tw//handle/246246/275687立體資料的視覺化(volume visualization) 是一個普遍被應用在科學計算和醫學影像上的技術，根據輸入資料本身的特性，則又可分為對於純量場(scalar field) 的立體渲染(volume rendering)；以及向量場中流場的視覺化(flow visualization) 等多種不同技巧。在本論文中，我們開發出三個強化立體渲染的技術；接著提出一個將向量場轉換成純量場的方法，使得我們的方法能被應用在流場的視覺化中。立體渲染的成像有三個關鍵議題，它們分別是：分類(classification)、打光(lighting)、和透明結構三維至二維的投影(3D to 2D projection)，如果沒有恰當地分類出令人關注的特徵，資料就無法正確被解讀；而如果沒有適當的打光，使用者就不能對資料擁有足夠的空間感；這兩者再加上從三維結構投射到二維影像所造成的維度縮減(dimension reduction)，造成歧異(ambiguities) 的現象。基於分類、打光、和感知(perception) 等技術，我們提出消除歧異的新方法，對於分類，我們結合體素(voxel) 的位置和強度後做維度縮減，產生二維的特徵空間，特徵空間與原始的體素強度所組合而成的三維轉換，可以對原始資料產生較好的群聚性(clustering)，使得不同的材質(materials) 可以簡單地被分類出來。對於打光，我們提出一個新的打光模型能即時地趨近多盞光源所產生的效果，這個模型可以容易地被延伸到切平面(plane cutting) 和最大強度投影(maximum intensity projection) 等方法上，讓使用者能洞察資料的內部結構；對於感知，我們則是利用視覺暫留的技巧配合立體成像(stereoscopic displays) 來強化透明立體資料的三維感知，並以使用者測試來驗證其效果。我們並且將這些技術推展到向量場的視覺化，以克服在三維空間下普遍面臨到的遮蔽(occlusion) 問題。相對於傳統上尋找流線(streamline) 的方法，我們統計粒子(particles)在體素間的相互流動(transitions)，再以馬可夫鍊(Markov chain) 將向量場轉化為純量場，經由實驗顯示，這個轉換可以描述出整個向量場的大致型態，並且可以搭配我們開發的立體渲染技術來消除向量場所形成的歧異。Volume visualization is a widely used technique for scientific computing and medical imaging. According to the different types of input data, several methods were developed for visualization, such as volume rendering for scalar fields and flow visualization for vector fields. In this thesis, we develop three approaches to improve the rendering results of volume rendering and then provide a method to transform vector data into scalar data so that our methods can be applied to flow visualization. There are three critical issues in volume rendering: classification, lighting, and 3D to 2D projection of transparent structures. Without proper classification to show interesting features, it is impossible to correctly interpret the volume data. Without lighting properly, users cannot gain sufficient spatial perception. In addition, volume rendering projects multiple transparent structures into an image plane and blends them together, so ambiguities are caused by these factors. Therefore, based on the classification, lighting and perception approaches, we propose novel methods to resolve the ambiguities. With classification, only voxels’ location and intensity are combined and reduced in dimensionality to form a 2D feature space. Augmented with intensity, the new 3D space can generate a better clustering for original data such that different materials can be easily classified. For lighting, a new lighting model is proposed to interactively approximate the effect of multiple lights. This model can be easily extended to plane cutting and maximum intensity projection to allow users to view the interior of the volume better. With perception, a thaumatrope approach is implemented on a stereoscopic display to enhance spatial perception of transparent volume data and a user study was performed to verify the effectiveness. Moreover, these methods are utilized in the field of flow visualization for solving the generally faced occlusion problem in a 3D vector field. In contrast to the traditional methods based on streamline tracking, we static the transitions of all voxles and then transform the vector field to a scalar field using a Markov chain. The experiments show that the clustered result makes a brief description of the original vector field and can co-operate with our volume rendering methods for disambiguation.18454341 bytesapplication/pdf論文公開時間: 2015/7/20論文使用權限: 同意有償授權(權利金給回饋學校)視算積體成像流場視覺化volume visualizationvolume renderingflow visualizationthesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/275687/1/ntu-104-D98944007-1.pdf