Feed and fly control of visual scanpaths for foveation image processing

Foveation-based processing and communication systems can exploit a more efficient representation of im- ages and videos by removing or reducing visual information redundancy, provided that the sequence of foveation points, the visual scanpath, can be determined. However, one point that is neglected by the great ma- jority of foveation models is the ”noisy” variation of the random visual exploration exhibited by different observers when viewing the same scene, or even by the same sub- ject along different trials. Here a model for the generation and control of scanpaths that accounts for such issue is pre- sented. In the model, the sequence of fixations and gaze shifts is controlled by a saliency-based, information forag- ing mechanism implemented through a dynamical system switching between two states, ”Feed” and ”Fly”. Results of the simulations are compared with experimental data derived from publicly available datasets.