Buiatti, Marco; (2002) Adaptation of the visual system to the temporal statistics of natural images. Masters thesis (M.Phil), UCL (University College London).

Text Adaptation_of_the_visual_syste.pdf Download (7MB)

Abstract

This thesis focuses on ways of dynamically adapting the visual system to the temporal statistics of natural images. Recent experiments suggest that adaptation relies on a coding strategy that has the effect of normalising the neuronal activity with respect to the standard deviation of the stimulus. We investigate the hypothesis that such a nonlinear coding strategy could represent a key mechanism in natural vision both for adaptation and efficient representation of the stimulus. We model the process of variance normalisation in a simple way, and we train it on time series mimicking the typical visual input of a human photoreceptor in a natural environment. Simulations confirm our hypothesis: variance normalisation adapts the wide natural range of light intensities into the limited neuronal one and efficiently removes almost all the redundancy present in the temporal structure of natural images. Moreover, despite the long-range correlations present in the natural input, the integration time corresponding to the optimal normalisation is surprisingly short and compatible with the one observed experimentally. The importance of this result seems to lie in the simplicity of the model. In order to prove its computational power, it is shown that it efficiently removes the redundancy of time series of very different nature - financial time series - also characterised by long-range correlations. This result suggests that such a method efficiently removes the redundancy of a wide variety of data, no matter how much correlated they are.

Download activity - last month

Export as JSONCSVXML

Download activity - last 12 months

Export as XMLCSVJSON

Downloads by country - last 12 months

Export as JSONCSVXML

Archive Staff Only

View Item