a

The measure and analysis of eye movements is crucial to neuroscience and psychology [10]. Eye movements are extremely useful from a methodological point of view- among other things, it is relatively easy to train animals to respond using eye movements, and the neurophysiological path-ways involved are relatively well understood [6]. Eye movements are also tremendously interesting as an object of study in their own right, because they are the most immediate means we have to explore our visual environment. In humans, the eyes do not move constantly but rather alternate between periods of relative stability, called fixations, and periods of movement. Very often the analysis is not concerned with movements but rather with fixations, and most especially where fixations occur. For example, in so-called free-viewing experiments, subjects view natural images, with no specific instructions-they are free to look wherever they like. Where they choose to fixate is far from random: subjects focus on similar locations, and exactly why they do that is an old debate in neuroscience and psychology [11, 8]. Some authors have argued that eye movements are controlled by a cortical saliency map [5], which represents interesting locations in the visual field, and that interestingness is computed very ha l-