Adaptation of image visual search strategies to utility given task and perceptual constraints

How do people adapt search strategies for finding visual images? An assumption in studies of rational behaviours is that the choice of strategies is shapec by rewards. Vision studies also find humans have a tendency to shift attention to the location that maximises information gain. However, it is unclear whether people would maximise information gain in circumstances where the cost of gaining information is likely to reduce its utility. In addition, information maximisation theories, derived from studies of a single saccade or search for a well-defined target with sequential optimal saccades, are unlikely to lead directly to an understanding of the bounds on visual search mechanisms in more complex and natural visual search settings, it is not known, for example, how people adapt basic elements of eye novement performance to changes in expected utility. Evidence from Experiments 1-4 showed that the selected strategy is sensitive to the display layout, i.e. the density and shape. Experiment 4 also observed that the choice of search strategy is associated with the expected number of alternatives in the search set. The follow-up diffusion based model showed how strategic control of thresholds for target and distractor recognition, in a gaze decision, could lead to the observed overall search behaviour in which participants balance the trade-off between the gaze duration and number of gazes. It also showed how saccade criteria are adapted to the number of alternatives. The findings of Experiment 5 supported the hypothesis that the micro-structure of the visual search process is adapted to a desire to maximise utility. The results showed that participants can trade time for value and that they can adjust gaze duration, and targeting, to the distribution of reward. The follow-up signal detection based model successfully predicted that adaptation generates optimal utility given the human visual and task constraints.EThOS - Electronic Theses Online ServiceGBUnited Kingdo