An Investigation of Complex Visual Search With The Genetic Algorithm

Abstract

Visual search is an everyday task, defined as the search for a target amongst distractors. Visual search has historically been thought to reflect primarily early perceptual processes, with little influence of cognitive processes. However, recent research suggests that this is an artefact of the simple stimuli typically used in the field. In this thesis, I will use the genetic algorithm to study visual search for complex stimuli. In part 1, I use the genetic algorithm to investigate a display comprised of 119 distractors, each with an orientation, colour and size. Results suggest that distractors differing in colour but sharing orientation with the target facilitated performance. This result is not predicted by common visual search models, posing an external validity problem for experiments using simple stimuli. Part 2 examines the processes behind visual search for colour, using displays that took advantage of colour labelling. The results show that it is only possible to search optimally when the target can be easily encoded in language. Further experimentation shows that priming a colour target that cannot be easily encoded in language then allows for optimal search, indicating that a top-down mechanism not based in language is used in search for colour. In part 3, we study linear separability, where search for a target is difficult if it is presented with distractors with more of a feature and other distractors with less of that same feature. We systematically varied the orientation of the target and the distractors, finding two independent measures of performance. One measure was correlated with bottom-up ability, while the other was correlated with the top-down ability. In the final chapter, I discuss the results of the three parts in the context of search for complex, heterogeneous displays. I speculate on the nature of the top-down mechanism, and offer directions for future research.