Active Object Recognition with a Space-Variant Retina

Figure 8

A high-level depiction of a gnostic field for classifying nonhuman apes using our space-variant ICA filters. The model splits the ICA filter output into chromatic, achromatic, and combined channels. This visual information projects to a gnostic set for each category, with units in the gorilla gnostic set responding strongest. The output of each gnostic set is given by the most active unit, and subsequent competitive normalization adjusts the activity to suppress the output of the chimpanzee and orangutan sets. Finally, evidence from the current fixation is added to the model’s beliefs, and information from all categories and channels is combined using a linear classifier.