For decades, researchers have examined the elements that influence how attention is directed during visual tasks, but few studies attempted to quantify the weighting of multiple of these aspects inside and across tasks in order to better understand their relative contributions. For a study, they sought to investigate the contributions of saliency, center bias, target characteristics, and object identification uncertainty in predicting the first nine fixation changes performed during free viewing and visual search tasks in the OSIE and COCO-Search18 datasets, respectively.
By providing a novel approach to assessing uncertainty in a picture based on object recognition, they focus on the latter-most and least familiar of these elements. They propose that the more object categories contending for an item proposal, the more the confusion about how that object should be recognized and, hence, the higher the demand for attention to address this doubt. Target features, as expected, better predicted target-present search, with their dominance concealing the usage of other characteristics.
Surprisingly, they discovered that target characteristics were only employed seldom during the target-absent search. In addition, they discovered that object recognition uncertainty outperformed an unsupervised saliency model in predicting free-viewing fixations, despite saliency being marginally more predictive of search. They found that uncertainty in object recognition, a metric that is picture computable and highly interpretable, predicted attention better than bottom-up saliency during free viewing.