Abstract
Purpose: :
During fixation, retinal images jitter around the fixation point, partly independently in both eyes. Since we are not aware of the jitter (despite that the amplitudes should be large enough to be seen), it is assumed that the image is stabilized by neural processing. We have tested how sensitive random dot stereopsis is against independent random spatial jitter of the two patterns to be fused, to learn about the performance of the "image stabilizer".
Methods: :
Two random square random dot patterns with 4 deg angular extend in the visual field were alternatingly presented on top of each other at 30 Hz on a computer screen. A pink/green spectacle was used to make only one pattern visible for each eye. The two patterns contained a 2 deg squared random dot field in the middle that could be displaced with mirror symmetry in horizontal direction (Julesz 1964) by 4 min of arc, causing a strong impression that the square was in front or behind the surrounding pattern. The patterns for each eye could be independently and randomly jittered with different angular amplitudes. Furthermore, 5 different patterns pairs could be shown in rapid sequence at 30 Hz for each eye.
Results: :
Thirteen adult subjects could reliably tell whether the central square was in front or behind the reference plane, even with independent random spatial jitter of the patterns shown to either eye. At 0.53 deg average jitter amplitude, 77% of the response were still correct, even though the disparities were 8 times smaller than the jitter amplitudes and 16 times smaller than the dots in the patterns. This result could be explained by a post-receptor image stabilization in both eyes that operates at least at 30 Hz before the information from both eyes is combined in the cortex. A less likely explanation is that pattern matching occurs in fact at 30 Hz (even though their positions in the left and right eye were variable). To test this hypothesis, 5 different pattern pairs were exchanged at 30 Hz, and jittered in addition. Surprisingly, depth perception was still functional and not different from the results with a single pattern.
Conclusions: :
Random dot stereopsis operates extremely fast (30 Hz at least per pattern pair to be matched) and robust against jitter (retinal positions can randomly vary much more than the disparity to be decoded). This finding challenges the standard model of neural disparity coding where the neurons are assumed to have a fixed preferred disparity that originates from their binocular receptive fields. It confronts us with the intriguing question how invariance against random variations in the offset between the left and right image can be achieved by the neural circuitry and involves parallel analysis of all dots since one diopter of defocus blocks random dot stereopsis.
Keywords: eye movements • binocular vision/stereopsis • pattern vision