A Structural and Semantic Probabilistic Model for Matching and Representing a Set of Graphs

This article presents a structural and probabilistic framework for representing a class of attributed graphs with only one structure. The aim of this article is to define a new model, called Structurally-Defined Random Graphs. This structure keeps together statistical and structural information to increase the capacity of the model to discern between attributed graphs within or outside the class. Moreover, we define the match probability of an attributed graph respect to our model that can be used as a dissimilarity measure. Our model has the advantage that does not incorporate application dependent parameters such as edition costs. The experimental validation on a TC-15 database shows that our model obtains higher recognition results, when there is moderate variability of the class elements, than several structural matching algorithms. Indeed in our model fewer comparisons are needed.