Abstract
We introduce a new descriptor for images which allows the construction of efficient and compact classifiers with good accuracy on ob ject category recognition. The descriptor is the output of a large num- ber of weakly trained ob ject category classifiers on the image. The trained categories are selected from an ontology of visual concepts, but the in- tention is not to encode an explicit decomposition of the scene. Rather, we accept that existing ob ject category classifiers often encode not the category per se but ancillary image characteristics; and that these ancil- lary characteristics can combine to represent visual classes unrelated to the constituent categories’ semantic meanings. The advantage of this descriptor is that it allows ob ject-category queries to be made against image databases using efficient classifiers (ef- ficient at test time) such as linear support vector machines, and allows these queries to be for novel categories. Even when the representation is reduced to 200 bytes per image, classification accuracy on ob ject cat- egory recognition is comparable with the state of the art (36% versus 42%), but at orders of magnitude lower computational cost.