Automated 3-D Feature Extraction from 3-D Data Sets

本研究的目的是将二维和三维稳态和二维随时间变化的流体流动数据集分解为基本结构，以便于描述、分析和比较。从数值计算和现代（光学）诊断中获得的结果往往过于复杂，无法进行人工检查、操作和显示。为了便于理解数据，需要对这些结果进行更简单但仍然准确的描述。将开发新的方法来从矢量数据中自动提取特征。矢量场，如速度、涡度或压力梯度，将被分解成基本结构，如临界点和划分流线。识别拓扑特征并确定它们之间的联系。特征和连接集以图的形式表示，可以从中选择和显示拓扑上重要的特征。这种简化的特性描述还有助于数据解释，并允许使用语法模式识别在数据库之间进行比较。这些数据可能是由于大规模计算或多通道实验测量。随着新型光学三维流诊断技术的发展和超级计算机速度和存储能力的提高，对这一研究的需求日益迫切。目前开发的数字图像处理和模式识别算法主要针对二维图像。将这些技术扩展到更高维度的数据库需要仔细考虑，这不是一项简单的任务。研究似乎有希望开发自动方法来操纵，理解和显示多维向量场将进行。

The objective of the research is to decompose two-dimensional and three-dimensional steady state and two-dimensional time dependent fluid flow data sets into elementary structures for purposes of description, analysis and comparison. Results obtained from numerical calculations and modern (optical) diagnostics are often too complicated for manual inspection, manipulation and display. A simpler but still accurate description of these results is needed to facilitate data understanding. New methods will be developed for automatic extraction of features from vector data. Vector fields such as velocity, vorticity or pressure gradient are to be decomposed into elementary structures such as critical points and dividing streamlines. Topological features are recognized and the connections between them determined. The set of features and connections are represented in the form of a graph, from which displays of topologically significant features can be selected and displayed. This simplified feature description also aids data interpretation and allows comparison between data bases using syntactic pattern recognition. Such data can be both due to large- scale computations or multi-channel experimental measurements. The need for this research is becoming increasingly urgent with the development of new optical three-dimensional flow diagnostic techniques and improved speed and storage capability of supercomputers. Currently developed digital image processing and pattern recognition algorithms are mainly aimed at two-dimensional imagery. Extension of these techniques to higher dimensional data bases needs to be carefully considered and is not a simple task. Research that appears to hold promise for developing automatic methods for the manipulation, understanding and display of multi-dimensional vector fields will be carried out.