Unsupervised Visual Object Categorisation with BoF and Spatial Matching

摘要: The ultimate challenge of image categorisation is unsupervised object discovery, where the selection categories and assignments given images to these are performed automatically. setting prohibits use best discriminative methods, in Tuytelaars et al. [30] standard Bag-of-Features (BoF) approach best. downside BoF that it omits spatial information local features. In this work, we propose a novel method which uses find initial matches for each (pre-filter) then refines ranks them using matching Unsupervised visual discovery by normalised cuts algorithm produces clusterings from similarity matrix representing match scores. our experiments, proposed outperforms al with Caltech-101, randomised Caltech-256 data sets. Especially large number classes, clear statistically significant improvements achieved.