Barbeit (Bachelor's Thesis)

Quantitative comparison of stable representations of natural stereo-images and binocular complex cells


Complex cells in primary visual cortex exhibit particular spatial properties such as tuning for stimulus orientation and spatial frequency independent of precise stimulus position in the receptive field. Recently it has been shown that these neurons share important such properties with simulated cells, which were adapted to exhibit optimally stable activity to natural visual stimuli. Consequently, it has been suggested that complex cells can be described as forming optimally stable representations of their natural input.
The simulation from this study has now been extended into the 3D domain by optimising activities of simulated cells to binocular cat-cam video sequences of natural scenes and in the work at hand I compare properties of the resulting stable cells to those of complex cells in order to evaluate whether a similar conclusion as from the simulation with monocular stimuli can be drawn. Thereto I directly analyse subunit receptive fields, examine binocular interaction profiles and investigate responses to random-dot stereograms while mainly contrasting my findings to physiological results obtained by the groups of Ralph Freeman and Andrew Parker.
Similar to the monocular study I find that simulated and real binocular neurons have many properties in common. Stable cells again exhibit tuning for orientation and spatial frequency as well as position invariance. Furthermore, all of the simulated cells are disparity selective and their disparity tuning curves show characteristics comparable to those of striatal neurons. Moreover, a predominance of phase encoding can also be observed in both systems.
Nevertheless, I notice several differences between stable and complex cells. Many of them are due to the limitations of the cell model employed in the simulation, which for example does not allow consistent inhibitory input from one eye like it is found in some real neurons. One crucial discrepancy, which I cannot explain, is a greater number of stable cells preferring large phase disparities near $\pm\pi$. I propose that this may result from properties of the used natural stimuli.
In conclusion, I find striking similarities between stable and complex cells signifying that binocular complex cells can be described as forming optimally stable representations of natural visual input as well. However, the cell model should be adapted to incorporate recent advances in the modelling of striatal neurons and further investigations need to be done to explain open discrepancies concerning phase disparity.

Und hier das ganze Ding zum Downloaden mit Bildern und bunt!: barbeit.pdf (1.9 MB)
Quelltext?: barbeit.tex (117 kB)