TOP  >  Thesis/Dissertation  >  {sotsurontitle}

Visualization of Multi-Attribute Data based on Drawing Density

Data visualization techniques are important in various fields of natural sciences and social sciences. In this study, we propose a new visualization technique for multi-attribute numerical data with three-dimensional grid structure. Three-dimensional grid of numerical data is widely used in the field of natural science, e.g. MRI data, geological data, meteorological data for air, and marine meteorological data. We focus on visualizing marine meteorological data. When we visualize such data, the most important point is how to make us recognize the characteristics of natural phenomena from the data. While many methods for visualizing single-attribute data have been proposed. They are not enough for analyzing natural phenomena from multi-attribute data. This is because, in many cases, characteristics of such natural phenomena are caused by mutual effects between various attributes. Various methods for visualizing multiple attributes simultaneously have been studied in recent years. They use three-dimensional graphics for visualizing multiple attributes and users can analyze characteristics of natural phenomena. However, visualization with three-dimensional graphics has an occlusion problem; a problem that some elements generated by visualizing a single-attribute value occlude other elements. Some methods solve this occlusion problem by arranging the position of all of those elements for avoiding the occlusion. However those methods are not suitable for visualizing a data whose physical layout of attribute values has important information for understanding data. We propose a new visualization method with two-dimensional graphics to solve the occlusion problem. The basic idea of our method is reducing the number of elements which make occlusion on the image. However, reducing the number of elements causes a problem of losing the amount of information visualized by those elements. To solve the problem of loss of information, we focus on two points. Firstly, occlusion occurs in limited areas of images generated by visualization. Secondly, if there are some areas of a visualized attribute value which is not occluded by elements of visualized other attribute value nearby a occluded area, human vision can interpolate what is in the occluded area. So, it is enough to reduce elements causing to occlude from only the areas where the elements occlude others extensively. Our method is composed of three processes described below. Firstly, we generate some images of each visualized single-attribute values of multi-attribute data with different visualizing methods. Then, these images are merged into an image, and divide the image into small regions. Secondly, we define gDrawing Densityh as quantifying the extent to which the occlusion has occurred in each small region. Finally, in just some region in which drawing density exceeds the threshold, we generate a new image of visualized single-attribute values which causes to occlude other attribute value. Then, we merge again the image with images of visualized other single-attribute values. We carried out experiments that compare how accurately human can read information from images visualized from marine meteorological data between with proposed method and without proposal method. Firstly, there are two groups which testers belong to, and we showed one by each group two images which are part of images of visualized multi-attribute data. One of two images is applied the proposed method, and the other is not. Secondly, testers drew graphs about each attribute value on a line which images had. As a result, by applying the proposed method, it was partially confirmed that there improve the accuracy of reading of occluded attribute values.