Structural Indexing Extended to Fuzzy Graphs of 2D Parts

Robust object recognition and image retrieval often lead to the construction of attributed graphs. When dealing with images of real scenes, these graphs must include fuzzy attributes to account for interpretation uncertainties and describe properly the observed object or scene. This paper first presents a 3D object model that requires the use of a graph with fuzzy attributes. This model is computed from a single 2D image of an object. Then, our method to match these graphs is introduced. This method is an extension of structural indexing. Being based on structural indexing, our method can match graphs with fuzzy attributes in large databases. A first experiment demonstrates the validity of our matching method, and a second one compares the performance of the combination of our model and our matching method with humans performing a similar task..