The approach presented here provides several advantages: first, the use of monkeys additionally allows the recording of single neurons (cmp. Maldonado et al., 2008). Second, it presents a tool to classify fixations that enables to relate neuronal activity to natural behavior (see Discussion), without making assumptions find more about the meaning

of the images to the observer. Third, our approach can be generalized to eye movements of humans. We find that in most cases, the subjective ROIs match well both the objects in the scene and the ROIs defined by their saliency maps. Exceptions are scenes containing human or primate faces. We made use of a Markov chain (MC) analysis to investigate the sequences of visited ROIs (assumed to be the states of a random walk) and extract their probabilities. Our approach of the scanpath analysis differs from Feng (2006) (reading task experiment), Van Der Lans et al. (2008) (search task), and Simola et al. (2008) (word search task) in that we feed the MC algorithm with the extracted ROIs. Such an investigation of fixation sequences shows that during free viewing of natural scenes a fixation is most likely to occur within the same ROI where the previous fixation occurred, Selleck GDC 0199 suggesting that local object exploration is executed before directing the focus to a new ROI. Three monkeys (D, M, and S) participated in

an electro-physiological experiment over many sessions, in which they were exposed to different natural images for 3–5 s, interleaved ifenprodil with blank screens or blank screens with a fixation spot (see Fig. 1, and Section 4.1 for details). Their eye movements were recorded with a scleral search coil, while the animals were allowed to freely explore the monitor screen with self-initiated eye movements (see Fig. 2A as an example of one image overlaid by an exemplary scanpath and the respective

fixations). An automatic algorithm extracted the fixations and saccades performed by the monkeys from the vertical and horizontal eye movements (Fig. 2B, see Section 4.2. for details), and derived the distributions of fixation and saccade durations (Figs. 2C, D). The distributions of fixation durations derived from all sessions and for all images (Fig. 2C) of monkeys D and M have very similar shapes, the mean fixation durations being 310 ms and 240 ms, respectively. These values correspond well to average fixation durations reported for humans during exploration of natural scenes, found to be in the range between 260 and 330 ms (Castelhano and Henderson, 2007 and Ossandon et al., 2010). However, the distribution of fixation durations of monkey S (Fig. 2C, red) differs from the distributions of the two other monkeys: it is broader, less skewed and has a heavy tail, and exhibits a much longer mean fixation duration (420 ms).