Limits to visual processing: non-linearities, noise and integration

点击翻译按钮获取中文摘要

In any sensory system there are three fundamental stages in processing; the initial filtering, non-linear transduction and integration. Neural signals are perturbed by internal noise that limits detectability. Previous approaches have often assumed the visual system is linear and the neural noise is additive. However, it is now recognized that the human visual system has nonlinearities that also affect these measurements. Furthermore, differences in this nonlinearity can be confused with differences in internal noise level if these effects are not accounted for. These effects require a new comprehensive approach that will distinguish nonlinearities from internal noise. In this proposal I develop novel methods that will have long term applicability for understanding the functional limits of visual processing in the cortex. This will be applied in the short term to contrast processing, and to the processing of form and motion in the intermediate term. I will principally use psychophysical methods for my investigations. These are carefully-controlled behavioural experiments which, when combined with computational modelling, allow us to investigate how the visual system processes its input in order to drive human behaviour. My investigations into the non-linearities present will be based on the well-established pedestal masking paradigm. This method measures the derivative of any non-linear transformation that is applied to the input signal in the brain. I also propose a novel paradigm that will allow one to measure the range over which neural noise is integrated at the various brain areas relating to different stimulus features. The short-term objectives are to provide a complete understanding of the limits to contrast detection in normal healthy observers and how contrast inputs from early visual areas are integrated across space to allow for the global processing of contrast. For each of the dimensions of spatial frequency, temporal frequency, chromaticity, field eccentricity and mean luminance. I will determine the relative limitations from non-linear transduction, internal noise, and spatiotemporal integration. This will serve as a foundation for subsequent work on how each of these fundamental limitations changes with age in healthy observers and with disease in patient populations. The longer term objective of our program of research is to investigate, how three fundamental factors nonlinearities, noise sources, and spatiotemporal integration contribute to the detection of stimulus features from contrast to motion, contours, shape, shape-from-motion, optic flow and stereo, both for local computations typically carried out in striate cortex as well as more global computations typical of extra-striate cortex. The approach taken here is both fundamental and novel. It involves a principled model-based framework in which to answer some of the basic questions in vision