Human Visual Search Follows Suboptimal Bayesian Strategy Revealed by a Spatiotemporal Computational Model

Humans perform sequences of eye movements to search for a target in complex environment, but the efficiency of human search strategy is still controversial. Previous studies showed that humans can optimally integrate information across fixations and determine the next fixation location. However, their models ignored the temporal control of eye movement, ignored the limited human memory capacity, and the model prediction did not agree with details of human eye movement metrics well. Here, we measured the temporal course of human visibility map and recorded the eye movements of human subjects performing a visual search task. We further built a continuous-time eye movement model which considered saccadic inaccuracy, saccadic bias, and memory constraints in the visual system. This model agreed with many spatial and temporal properties of human eye movements, and showed several similar statistical dependencies between successive eye movements. In addition, our model also predicted that the human saccade decision is shaped by a memory capacity of around 8 recent fixations. These results suggest that human visual search strategy is not strictly optimal in the sense of fully utilizing the visibility map, but instead tries to balance between search performance and the costs to perform the task.