High-dimensional Spectral Feature Selection for 3D Object Recognition based on Reeb Graphs

In this work we evaluate purely structural graph measures for 3D object classification. We extract spectral features from different Reeb graph representations and successfully deal with a multi-class problem. We use an information-theoretic filter for feature selection. We show experimentally that a small change in the order of selection has a significant impact on the classification performance and we study the impact of the precision of the selection criterion. A detailed analysis of the feature participation during the selection process helps us to draw conclusions about which spectral features are most important for the classification problem.