Embedded Files
Background
Background
Everyday you face the challenge of finding or detecting an item you are looking for. For example, before you leave the house you have to look for your keys on the kitchen table. "Visual search" is the term psychologists use to describe this situation, in which you are locating or detecting a specific item using your vision. We call the specific item you are looking for the "target", and any other nearby items "distractors". For over 50 years we have tried to understand the processes which take place when we perform visual search. We would like to understand the steps and computations used by the brain to go from the image you receive through your eyes, to a decision.
Often when we do visual search we receive incomplete or uncertain information, and there may be lots of distractor items. For example, you might only make a quick glance for your keys on the kitchen table, and the table may be covered in other objects! Some of the distractor items might look like the target, your keys. For example a bottle opener might have a somewhat similar shape, and also be metallic.
Several theories try to explain how our brains deal with incomplete and uncertain information. A mathematical formula called Bayes rule describes the optimal way to use incomplete and uncertain information. In a glance, you might not get a good look at any object on the kitchen table, but you will get some information, such as the rough shape of the objects, and an estimate of their colour. Bayes rule tells us the best way to combine this information to arrive at an estimate of whether the keys are on the table. Work by other researchers has suggested that our behaviour closely matches what would be expected if our brains were using Bayes rule.
This study
This study
We wanted to look at a specific aspect of visual search, the effect of distractor items. Distractor items are all the items which are not the target, so they are everything on the kitchen table which are not your keys. Previous research has suggested that distractors affect the difficulty of visual search. If the distractor items are all very different to each other (the technical term is "high variance") then visual search becomes more difficult. Visual search becomes easier if the distractors are very different to the target (the technical term is "high mean"). We wanted to see if we could find these effects, and whether we could explain them using the idea that our brains use Bayes rule.
In the study, participants were briefly presented with a display of striped patches (equivalent to glancing at the table when looking for your keys). Participants had to decide whether or not there was a striped patch at 45 degrees clockwise from vertical (this target patch is the equivalent of your keys).
The task that participants completed.
Using the idea that our brains use Bayes rule, we built a model which predicts how people will respond when presented with a certain set of patches. We then looked to see if the model can match the data that was actually collected. In the diagram we have the real data, shown using capped bars, and the fitted model shown using shading. On the x-axis we have plotted how different the distractors are from each other (distractor variance), and the different colours represent how different the distractors are from the target (distractor mean). The three different plots have different y-axes. In the top plot, the y-axis is accuracy, so this plot shows how accuracy changes with distractor variance. In the second plot, the y-axis shows hit rate, which is the proportion of times participants said that the target was present when it was in fact present. In the third plot, the y-axis shows false alarm rate, which is the proportion of times participants said that the target was present when it was in fact absent.
As you can see, the model does a pretty good job of capturing the patterns in the data. This suggests that Bayesian models of visual search provide a good description of how distractors affect behaviour, in turn suggesting that they might be a good description of the steps and computations used by the brain. It should be noted that another model, based on a different idea to Bayes rule, could also fit the data well. As a result we can't be sure the brain is using Bayes rule, but the results do suggest that this and related approaches are on the right track.
Model predictions (shading), match up with the real data (capped bars).
Page updated
Google Sites
Report abuse