Figure 6: Suppressive features strongly increase the sparseness of neural responses.

(a) The fraction of sparseness that remains after suppressive features are removed is plotted as a distribution across the data set of V2 neurons. Sparseness is computed across all natural image inputs presented to a given V2 neuron. (b) Cross-orientation suppression increases the sparseness of neural responses. Difference in sparseness ratio between the model based on all relevant Gabor features compared that where orientation of suppressive Gabors have been set at random.