Graph-Based Object Class Discovery from Images with Multiple Objects

摘要

Discovering objects models from image database has gained much attention. Although the BoVW (Bag-of-Visual-Words) approach has succeeded for this research topic, Xia and Hancock pointed out the two drawbacks of the BoVW: (1) it does not represent the spatial cooccurrence between local features and (2) it is difficult to select proper vocabulary size in advance. To overcome these drawbacks, they propose a novel unsupervised graph-based object discovery algorithm. However, this algorithm assumes that one image contains only one object. This paper develops a new unsupervised graph-based object discovery algorithm that treats images with multiple objects. By clustering the local features without specifying the number of clusters, our algorithm does not have to decide the vocabulary size in advance. Next, it acquires object models as frequent subgraph structures defined by a set of co-occurring edges which describe the spatial relation between local features.