Active Object Recognition with a Space-Variant Retina

Table 1

Mean per-class accuracy on Caltech-256 using 50 training instances per class for each channel specific gnostic field, along with the multichannel approach that combines all three channels.

Achromatic	Chromatic	All	Multichannel
45.6±0.5	31.4±0.4	48.4±0.5	50.8±0.5