Study Of Edge Bundling Based On Edge Clustering

In information visualization, nodes are widely used to represent objects, and edges are widely used to represent the relationships between nodes/objects. The expression of edge is so clear and intuitive that edges widely exist in various information visualization techniques, such as network graphs, hierarchical trees and parallel coordinates. However, with the growing of data size, the scale of visual layout is also expanding. Tens of thousands of edges overwhelm the display and cause the visual clutter problem.Many methods have been proposed to address the visual clutter problem. Among them, the edge-bundling method has become popular in recent years, because this method can bundle similar edges by compressing the spatial distance between edges and thus reduce the visual clutter, but without reducing the numbers of nodes and edges. However, most edge-bundling methods mainly focus on addressing the visual clutter problem and pay little attention to the accuracy of the bundling result, the level of data expression and the time efficiency of the program. Therefore, we propose the clustering-based edge-bundling method with the advantages of high time efficiency and high precision. The content of this paper includes the following three parts:(1) Firstly, the edge-clustering method is studied. The accuracy of edge-bundling results depends on the definition of edge similarity, because many edge-bundling methods group edges by clustering methods and the clustering methods cluster edges based on the edge similarity defined by a distance function. Therefore, we discuss three distance functions and three clustering algorithms in section 2.(2) Based on edge clustering, we further propose a fast edge-bundling method designed for large scale parallel coordinate plots. Our method firstly groups edges through edge clustering, and then calculates the mean value of each edge set to get a path. Edges are further bundled based on these paths. Finally, semantic enhancement is used to improve the visualization results.(3) After comparing edge-bundling methods for network graphs, we propose a new method based on the edge clustering and edge skeleton to improve the data expression ability and the algorithm efficiency. The new method clusters and bundles edges using the spatial distance between edges and the skeleton path, and it can effectively reduce the iteration times, simplify the process and improve the time efficiency.