| In this thesis, we venture into the realm of low-level vision modelling. The proposed model is unique in its ability to draw from fundamental engineering principles to explain the development, function, and organisation of cortical cells. We revert to nature's affinity for optimisation, relying on the eigenvectors of natural images as the central component of this model. We assert that the receptive field arrangements of the cells derive from the eigenstructure of images. We show that the standard features of orientation selectivity maps and structure arise through lateral competition in a self-organising map. We liken the image processing sequence in low-level vision to a decomposition of the eigenstructure of the image data. These results are achieved through the fusion of McGuire and D'Eleuterio's [1999] eigenpaxel algorithm, a localised principal component analysis technique, and Kohonen's self-organising map neural network structure. |
