Society for Neuroscience 2006
Oct. 14-18, 2006 in Atlanta, Georgia
Integration of spatial frequency channels for stereo correspondence in macaque area V4
Hironori Kumano, Seiji Tanabe, Ichiro Fujita
Graduate School of Engineering Science, Osaka University, Toyonaka, Japan
At an early stage of stereo processing, such as in area V1, neurons respond as if stereoscopic depth is encoded, even for stereograms that are binocularly anticorrelated. They also respond to grating patterns in a manner such that the encoded depth varies with the spatial frequency. These characteristics are consistent with disparity energy computations. To solve the correspondence problem, this rudimentary representation of stereoscopic depth needs to be further processed. Theoretical studies suggest that the stereoscopic system may take advantage of spatial frequency channels and integrate them to derive a neural representation that solves the stereo correspondence problem. We tested in area V4 whether the integration of multiple spatial frequency channels contributes to the attenuation of disparity selectivity to anticorrelated stereograms. We recorded both the disparity tuning and the spatial frequency tuning of single neurons in area V4 of two awake, fixating monkeys (Macaca fuscata). For testing the disparity tuning, we used stimulus patterns of either narrow-band or broad-band spatial frequency. The broad-band pattern was a random-dot stereogram (RDS) that was either binocularly correlated or anticorrelated. Across the population of neurons (N = 66), neurons with broader spatial frequency tuning had more attenuated disparity tuning for anticorrelated RDS. Examination of disparity tuning with narrow-band noise stereograms revealed that the preferred disparity of the majority of V4 neurons (47/72) was consistent across a range of spatial frequencies. Disparity energy models that include a phase disparity between the receptive fields of the two eyes do not predict this independence of preferred disparity to the spatial frequency of the stimulus. We suggest that V4 neurons pool disparity energy signals across spatial frequency channels for stereo correspondence computation.
Supported by grants from MEXT (17022025) and the Takeda Science Foundation.