Human search for a target on a textured background is consistent with a stochastic model

Area of Science:

• Cognitive psychology
• Computational neuroscience
• Visual perception

Background:

• Previous research indicated human visual search aligns with optimal strategies in spatial distribution and fixation counts.
• However, recent studies show failures of optimality in human saccade tasks.

Purpose of the Study:

• To compare human performance on a challenging visual search task with predictions from a stochastic search model.
• To investigate the spatial distribution of fixations and the number of fixations required by human observers.

Main Methods:

• A challenging visual search task was designed.
• A stochastic search model was developed, incorporating a target-visibility map from human detection performance.
• The model's saccade selection mimicked human saccade distributions when the target was not detected.
• Human observer performance (fixation counts and locations) was compared to model predictions.

Main Results:

• A memoryless stochastic model accurately predicted human performance in terms of fixation counts.
• Human fixation distributions deviated from the ideal observer's predicted doughnut shape, showing a central bias instead.
• The similarity between human and ideal observer fixation distributions did not replicate.

Conclusions:

• Human visual search in noise employs a near-random strategy, not an ideal one.
• Near-optimal performance is achieved due to inherent biases in human saccade distributions (central bias).
• Findings reconcile efficient search performance with observed suboptimal saccade behaviors, explaining discrepancies in prior research.