(Note, that for the latter analyses the background state was discarded.) CHIR 99021 This also holds true separately for monkeys D and M (Wilcoxon test, p < 0.001). As for saccade durations (Fig. 2D), the distributions of saccade lengths are skewed, thus showing a tendency for shorter with respect to longer saccades. In order to avoid any bias due to the skewness of the distribution, we performed

a second test, which, instead of uniform probabilities, took into account the actual saccade amplitude distributions. The expected transitions were weighted by the actual probabilities of saccade amplitudes (see Experimental procedures, Section 4.7 for details). The results confirmed the previous analysis, i.e., a significant larger probability of staying within a cluster and a significant lower probability of switching between clusters than expected SCH727965 (Figs. 5D, E in green). Overall, the Markov chain analysis revealed that the monkeys preferentially move their eyes within the same ROI before saccading out to another ROI or to the background. These results did not show any dependence on the contents of the images, in particular with respect to primate faces. Thus, the viewing strategy of the monkeys seems to be composed of sequences of local explorations of regions-of-interest, but not of random eye movements between ROIs. The present work shows that during free viewing of natural images, Cebus monkeys follow

a strategy that involves periods of local exploration, characterized by consecutive fixations that stay inside the same regions-of-interest. These periods of local exploration are typically followed by longer saccades into

a new ROI, where a new period of local exploration begins. ROIs were defined as areas containing clusters of Ureohydrolase fixations performed by the monkeys over several presentations of an image. For most of the images, the locations of the fixation clusters correlate well with saliency maps, suggesting low-level features as the driving force for the eye movements. Images containing faces are an exception, in that faces attract most of the fixations despite their very low saliency. Therefore, as hypothesized, subjective ROIs reflect both bottom–up and top–down factors. Our approach based on subjective ROIs is robust with respect to content and semantic meaning of the images, because it relies on the spontaneous sequences of eye movements performed by the subject. Similar approaches have been used in humans, showing conserved clusters of fixations in the same image for different subjects ( Judd et al., 2009). Our analysis of eye movement sequences during free viewing is based on the finding that fixations are not evenly distributed on an image, but rather define clusters, on top of conspicuous objects. This was the case for two out of three subjects studied (monkeys D and M). However, the third monkey (S) used a completely different viewing strategy.