Scanpath Prediction via Utilizing Peripheral Information of the Human Visual System

Predicting the gaze patterns of the human eye when performing a visual search helps understand the mechanism of the human visual system. Most current scanpath models consider the physiological properties of human visual space when modeling gaze behaviors, and use image patches with different resolutions to represent the central and peripheral vision. However, the simplified method of lowering the image resolution to simulate the peripheral scene cannot accurately represent the visual information loss. On the other hand, few studies have focused on how peripheral information loss affects human attention shift during visual search. To address the above issues, we propose a novel scanpath prediction approach by designing a Multi-scale Visual Representation Module (MVRM) to process peripheral information of the human visual system. Besides, we introduce a Neural Interaction Module (NIM) with three interaction mechanisms of visual signals to predict the next fixation point. Experimental results on the benchmark datasets demonstrate the superiority of our method.