Strahler based graph clustering using convolution

We propose a method for the visualization of large graphs. Our approach is based on the calculation of a density function resulting from the application of a metric on the vertices of a graph. The density function is then filtered using a convolution, leading to a partition of the graph. The choice of an appropriate kernel for the convolution makes it possible to control the number of clusters, and their size. Our algorithm can be executed automatically, but the parameters can also be interactively fixed by the user. We applied the algorithm to the problem of legacy code extraction from inclusion relation of C++ source files and film sequence analysis. The metric used here is defined from Strahler numbers, which measure the "ramification" level of graph vertices.