Modeling contextual influences on object recognition