WHO:	Ben Tatler , University of Sussex, UK
TOPIC:	Visual determinants of eye movements revealed by reverse correlation
ABSTRACT:	To determine the visual characteristics that specify where we move our eyes, we recorded saccades while viewing natural scenes. We then used the reverse correlation technique to determine the optimal (least squares) linear filter that operates on (potentially non-linearly transformed) images to generate the observed saccades. This method has previously been used to identify visual, auditory and tactile characteristics that determine spike generation in neurons. Here we attempt to predict saccaded-to locations. First we constructed saliency maps that made specific features of the images explicit: raw luminosity, difference of luminance from the mean, local contrast, chromaticity, and edge information based on the outputs of odd-symmetric Gabors. Each was performed at two spatial scales. We then measured saccade-triggered-averages (STAs) for each of these saliency maps. Unsurprisingly, no structure was found for the raw luminosity STA. For the other transformations, localised receptive fields were found at the high frequencies. STAs give some insight into the operation of the system, but a better characterisation is in terms of an optimal linear filter. The optimal filter is only equal to the STA if the inputs are uncorrelated but this was not the case. To calculate the optimal filter with correlated inputs requires a matrix inversion of the input covariance matrix, which can be numerically unstable. We therefore used ridge regression with cross-validation to identify these filters stably. Most interestingly, edge information was weighted positively at high frequencies, but negatively at low frequencies. We interpret this as eye movements being determined partly by surface property boundaries whilst being invariant to gross illumination changes.›
WHEN:	11/14/2003 12:00:00 PM
WHERE:	Meliora 269

Events Homepage