Semantic override of low-level features in image viewing - both initially and overall

Guidance of eye-movements in image viewing is believed to be controlled by
stimulus driven factors as well as viewer dependent higher level factors such as task
and memory. It is currently debated to what proportions these factors contribute
to gaze guidance, and also how they vary over time after image onset. Overall, the
unanimity regarding these issues is surprisingly low and there are results supporting
both types of factors as being dominant in eye-movement control under certain
conditions. We investigate in this paper how low, and high level factors influence eye
guidance by manipulating contrast statistics on images from three different semantic
categories and measure how this affects fixation selection. Our results show that the
degree to which contrast manipulations affect fixation selection heavily depends on
an image’s semantic content, and how this content is distributed over the image.
Over the three image categories, we found no systematic differences between contrast
and edge density at fixated location compared to control locations, neither during the
initial fixation nor over the whole time course of viewing. These results suggest that
cognitive factors easily can override low-level factors in fixation selection, even when
the viewing task is neutral.