Examination Committee

Prof Richard H Y SO, IELM/HKUST (Chairperson)
Prof Bertram E SHI, ECE/HKUST (Thesis Supervisor)
Prof Weichuan YU, ECE/HKUST
 

Abstract

The responses of disparity selective complex cells in the mammalian visual system are often modeled by the disparity energy model. This model linearly combines inputs from binocular simple cells, whose responses are computed by the combination of left and right eye inputs through linear receptive fields, followed by half squaring. However, many cells’ responses cannot be explained by this model. While this model has been extended to explain some specific types of tuning, actual neurons display richer characteristics than can be fully accounted for by these models.

Here, we describe a two layer simple cell model that can be used to construct complex cells that fully cover this range. The model combines the responses from a first layer of monocular simple cells. By adjusting the weights from the monocular to a second binocular layer, the model can exhibit more diverse tuning properties than previous models. We also show that these weights can be learned by sparse coding, and that if so, there is a strong relationship between distribution of tuning properties in the population and the input disparity statistics.

Our simulation results also show that the modeled population evolves towards the data recorded from cat Area 18, and we can find the optimal input disparity distribution that results in the most similar population with the recorded data. By changing the wavelength of model receptive field, we also illustrate that there is a linear correlation between the wavelength and the standard deviation of the optimal input disparity distribution.

Keywords- stereo disparity; disparity energy model; neural population; sparse coding; ocular dominance; disparity sensitivity