Gaze shifts are eye movements that play an important role: the Human Visual  System (HVS) achieves highest resolution in the fovea and the succession of  rapid eye movements (saccades) compensates the loss of visual acuity in the  periphery when looking at an ob ject or a scene that spans more than several  degrees in the observer’s field of view.

The ability to predict, given an image or a video, where a  human might fixate elements of a viewed scene has long been of interest  in the vision community.   Interestingly enough, one point that is not addressed by most models is one essential feature of gaze shifts: the  ”noisy”, idiosyncratic variation of the random exploration exhibited by different observers when viewing the same scene, or even by the same subject along  different trials. Such variations speak of the stochastic nature of scanpaths.

At the most general level one can assume any scanpath to be the result  of a random walk performed to visually explore the environment under the constraints of both the physical information provided by the stimuli (saliency or  conspicuity) and the internal state of the observer, shaped by cognitive task being involved) and emotional factors.

Apparently evolution has achieved efficient  eye movement strategies with minimal neural resources devoted to memory. This is in common with movement patterns and foraging behaviors of some biological organisms, e.g., spider monkey, albatrosses, jackals  that indeed can be represented as random trajectories  driven by alpha-stable distributions.

On such basis, in this talk,  a different view of the gaze-shift mechanism will be introduced  as that of a motor system implementation of an active random sampling  strategy  of the surrounding visual world.  Examples will be provided to show how such strategy can be exploited to carry on an attentive analysis of either static  and dynamic scenes.