Embedding labeled graphs into occurrence matrix

This paper presents an approach to the classification of structured data with graphs. We suggest to use a graph signature in order to solve the problem of complexity in measuring the distance between graphs. We choose a numerical matrix to embed both topological and labeling information. The use of this matrix allows us to use some classical tools of classification, computation of distances or feature selection. After a description of the matrix and the method to extract it, we compare the results achieved on public graph databases for the classification of symbols and letters using this graph signature.