Research on techniques for classifying and searching for large-scale volume datasets using critical point graphs

在本研究中，为了便于分类和搜索，我们开发了一种构造临界点图的技术，该临界点图可以用简单的几何图形表示体数据集。听着，我们把由计算机模拟和测量装置生成的数值数据集称为体积数据。2003年，我们开发了一种从标量体数据集构建临界点图的算法。当我们计算临界点图时，我们从距离临界点一小段距离的点沿着临界点特征向量的方向画一条向量线。由于目前我们还不知道多小的距离才足够，因此我们对临界点图的几何形状如何通过改变矢量线的起始位置而变化进行了灵敏度分析。2004年，我们开发了一种利用临界点图计算体数据集之间相似性度量的技术。首先，定义了临界点图的特征向量。特征向量由临界点信息及其连通性组成。前者包括临界点标识符、点坐标、标量数据值、临界点类型等。后者包括弧线标识符、两端的临界点和沿弧线的体积单元。我们假设临界点图是三维描述的特征，并开发了一种利用模式识别技术计算临界点图之间相似度的技术。

In this research, we developed a technique of constructing a critical point graph which can represent a volume dataset using a simple geometry in order to facilitate classifying and searching. Hear, we call a numerical dataset which is generated from computer simulation and measurement device as a volume data.In 2003, we developed an algorithm for constructing a critical point graph from a scalar volume dataset. When we calculate a critical point graph, we trace a vector line from a point which is located a small distance away from a critical point in the direction of the eigen vector of the critical point. Since, currently, we do not have any knowledge about how small distance is adequate, we carried out a sensitivity analysis about how a geometry of the critical point graph changes by changing a starting location of the vector line.In 2004, we developed a technique of calculating a similarity measurement between volume datasets using their critical point graphs. First, we defined a feature vector of a critical point graph. The feature vector is composed of information on the critical points and their connectivity. The former includes identifiers of critical points, point coordinates, scalar data values, types of critical points.. The latter includes identifiers of arcs, critical points at the both ends, and volume cells along the arcs. We assumed a critical point graph as a character described in 3D and developed a technique of calculating a similarity between critical point graphs by using pattern recognition techniques.