A Generic Engine for Recognizing Scene Categories

Because the trade-off between discriminative power and invariance varies from task to task, no single kind of feature is optimal in all situations. In this paper we build a generic scene recognition engine based on multiple cues which capture the different characteristics of the images. Our method can be equally applicable to a range of tasks including scene and object category recognition tasks. Experimental results show that the proposed algorithm outperforms the state-of-the-art methods on a large, challenging dataset of fifteen natural scene categories.