Spatio-temporal saliency model to predict eye movements in video free viewing

This paper presents a spatio-temporal saliency model that predicts eye movements. This biologically inspired model separated a video frame into two signals corresponding to the two main outputs of the retina (parvocellular and magnocellular outputs). Both signals are then decomposed into elementary feature maps by cortical-like filters. These feature maps are then used to form two saliency maps: a static one and a dynamic one. These maps are fused into a spatio-temporal saliency map. The model is evaluated by comparing the salient areas of each frame predicted by these saliency maps (static, dynamic, spatio-temporal) to the eye positions of different subjects during a video free viewing experiment with a large database (17000 frames).