Abstract
Direction discrimination thresholds are higher for motion along diagonal directions than for motion along cardinal directions. This has been shown both for human psychophysics and smooth pursuit (e.g., Krukowski & Stone, Neuron, 2005). The physiological mechanism underlying this 'oblique effect' remains unknown. Given the increasing use of 'oculometric' measures of motion perception (Stone et al., Perception and Eye Movements. In: Encyclopedia of Neuroscience, 2009), it would be useful to isolate the motion processing stage at which these perceptual and oculomotor direction biases arise. To this end, we tested whether elements of a model of visual motion processing based on V1 and Middle Temporal (MT) neurons (Perrone, JOV, 2012) could replicate the oblique effect. The input stimuli consisted of 128 x 128 pixel x 8 frame movies of a moving dot. We ran 50 trials at each of 5 dot direction angles (A ± 4° and ± 8°) where A = 90° (cardinal) or 135° (oblique). The final model output was derived from the weighted vector average of the individual local velocity vector estimates generated at frame 4 of the output. A linear regression of actual versus estimated direction produced a slope of 1.8 for the cardinal test and 0.6 for the oblique, thus replicating the typical empirical data pattern of higher gain for the cardinal versus oblique directions (Krukowski & Stone, 2005). In the model, systematic amplifications of direction space arise from the inhibitory mechanism used by Perrone (JOV, 2012, Fig. 9) to prune the direction of velocity outputs arising from the MT pattern units to some extent. These 'spurious' directional signals are inhibited by a signal from MT component units at the same location as the pattern units, before the output is fed into the velocity sensors. When this inhibitory signal was turned off in the model, the oblique effect disappeared.
Meeting abstract presented at VSS 2014