The human eyes observe and perceive different external scenes by the continuous movement of fixations. It is of great difficulty to predict and model "where to look" of human eyes in computer vision research. To address this problem

a computational model including two parallel processes: global transition and local movement

is proposed based on visual dynamic saliency. The former predicts the long distance migrations of saccades in the scene by combining the eye movement bias

visual centrality and inhibition of return; the latter is used to determine the next destination by searching for the local largest average salient point. And these two processes can be switched through a certain judgment rule. Experimental results show that the proposed model can more effectively simulate the human eye gaze point distribution

resulting in better performance.