Abstract
Given the importance of visual search for survival, humans may have evolved sophisticated fixation strategies for finding targets of interest quickly and accurately. We have derived the optimal search strategy for a visual system with variable sensitivity across its retina, where the task is to search for a target embedded in a noise background. The ideal search strategy is as follows: (1) during a fixation, collect all the available information (which is corrupted due to internal noise and sensitivity variation across the retina), (2) compute the (posterior) probabilities that the target is present at each potential location in the visual field, (3) given those posterior probabilities, compute the shortest sequence of subsequent fixations necessary to find the target with a desired level of accuracy. (4) execute the first saccade of this sequence, and (5) repeat steps 1 – 4. We have simulated an approximation to the ideal visual searcher and compared it with other less efficient searchers—making random fixations and fixating on the location with the highest posterior probability. At the time of writing this abstract we have not made detailed comparisons of human and ideal search patterns, but we have discovered some interesting properties of the ideal that are qualitatively similar to those of humans (and different from the less efficient searchers above). Specifically, even without time or energy costs for long saccades, the ideal searcher often (1) makes short- and intermediate-length saccades, (2) tends not to fixate near the edge of the display, and (3) sometimes fixates a central location within a cluster of locations with high posterior probability. The ideal visual searcher promises to serve several important roles: providing the appropriate benchmark for comparison with human performance; providing a starting point for developing real models of visual search; and helping to uncover the specific inefficiencies in human search performance.