Abstract:

PURPOSE: Figure–ground segmentation is a fundamental task for biological and artificial visual systems. The visual figure is many multiples greater in size than the receptive fields/filters of early visual areas. Computing figure-ground must therefore involve a cascade of bottom–up & top–down processes across multiple visual areas and spatial scales. We previously observed figure–ground enhancement at long latency (~100ms) at the earliest stage of central vision, the lateral geniculate nucleus (dLGN). Here we ask how large saccadic eye movements affect figure-ground modulation in dLGN. Critically, we wished to evaluate how figure-ground representations were affected across the saccade.
METHODS: We performed tetrode recordings of spiking and local field potentials (LFP) in behaving macaque dLGN. We used speeded reaction–time where the subject had to find, and then saccade to a moving-dot defined visual figure against a background. Spiking and LFP responses were analysed by aligned them to either the stimulus onset, or the target–saccade onset.
RESULTS: We found strong and robust saccadic modulation of neural responses in macaque dLGN across all locations analysed. Saccadic modulation was seen for parvo–, magno– and konio–cellular streams. Time-frequency analysis showed that saccadic modulation is largely driven in the α-band. In the trials where a figure stimulus was in the receptive field, a continued figure-ground specific enhancement was maintained even after the saccade terminated.
CONCLUSIONS: We found figure-ground selective enhancement is amplified across saccades (i.e. even after the stimulus has left the receptive field), showing that trans-saccadic continuity of representation may have significant implications for the stability of visual perception.

Key words: top-down, saccade, perception, figure-ground, LGN