Scene Classification via Hypergraph-Based Semantic Attributes Subnetworks Identification

Scene classification is an important issue in computer vision area. However, it is still a challenging problem due to the variability, ambiguity, and scale change that exist commonly in images. In this paper, we propose a novel hypergraph-based modeling that considers the higher-order relationship of semantic attributes in a scene and apply it to scene classification. By searching subnetworks on a hypergraph, we extract the interaction subnetworks of the semantic attributes that are optimized for classifying individual scene categories. In addition, we propose a method to aggregate the expression values of the member semantic attributes which belongs to the explored subnetworks using the transformation method via likelihood ratio based estimation. Intensive experiment shows that the discrimination power of the feature vector generated by the proposed method is better than the existing methods. Consequently, it is shown that the proposed method outperforms the conventional methods in the scene classification task.