Salience makes objects pop-up from their surrounding and immediately attract our attention. A salient object in the visual domain would, e.g., be a red dot among white ones. We can sometimes intuitively say an object is more salient than the other, e.g., a moving red dot might be more salient than a static one.

Is there a way to quantify how salient an object is?


To narrow down the question a little, suppose you have a set of different output devices that can provide either visual or auditory cues (LEDs, loudspeakers, lamps, screens...) and they are near a target object you which to attract attention to. Ideally one would choose the output device that is closest to the target object to attract attention, but depending on the distance, the output device may not be as conspicuous. So at a given distance, I would like to know, which device will attract attention the most. The environment would be mostly static and well lit.

  • 1
    $\begingroup$ The question is incredibly broad. It depends of the type of stimulus and the context (distractors) how salient something is. Visually speaking, with a static background a moving stimulus is likely more salient, but the same stimulus may be indistinguishable on a dynamic background. Could you please narrow down your question? Otherwise the answer would not become more satisfying than "subjective ratings of salience". $\endgroup$ – Robin Kramer Oct 7 '16 at 15:33
  • $\begingroup$ @RobinKramer I narrowed the question a little, do you know if some research has been done in this area? $\endgroup$ – Girauder Oct 7 '16 at 16:18
  • $\begingroup$ That is better already. There are still many things that may play a role but it is answerable I believe. My initial thoughts are that auditory stimuli can give some sort of emotional (i.e. startle) response, stronger than visual cues, and may therefore be more salient. However, the auditory sources are more difficult to locate compared to visual stimuli. I will see if I can find some references to back this up, and see if there are some measures of the salience, but other people are welcome to answer it as well :) $\endgroup$ – Robin Kramer Oct 9 '16 at 16:46
  • 1
    $\begingroup$ Girauder have you looked into signal detection theory? commenting because anecdotal but what I did to measure salience was a simple yes/no (2AFC) detection task and alter the target e.g. brightness/size. Then work out the threshold of detection from that - then compare this for the different targets. I'm not sure about auditory stimuli though. $\endgroup$ – queenslug Oct 10 '16 at 8:27
  • $\begingroup$ @queenslug I haven't looked into that yet, but sounds like it could be really helpful! Did you publish your research? I would love to read in more detail what you did! $\endgroup$ – Girauder Oct 10 '16 at 9:33

The duration of eye fixation has been used in both infants and adults to measure locus of attention. This can be achieved using an eye-tracker. A more primitive and less exact way is to manually count the frames of a video recording of subjects. The frames per second (FPS) would be a limiting factor. In addition, the item that was first observed is the more salient one, provided that there is no secondary bias in the task (i.e. the objects should be an equal distance apart or the distance could be randomized).

For visual search and serial search tasks, responses are gathered and individual response curves (reciever operator characteristics, ROC) for participants and conditions can be computed using signal detection theory (SDT). The discrimination index d prime (d') can then be calculated only if two explicit responses are required exist.

Another option is to vary the presentation time systematically in order to quantify processing speed, but this requires a more computational approach.

  • $\begingroup$ what do you mean with: the discrimination index d' can only be calculated if both match and non-match conditions exist? Thanks a lot for the info by the way. You mentioned "In addition, the item that was first observed is the more salient one" But what if one has 2 out devices that can be presented to a user, can one determine which one of the two will be more salient, or at least which one will be noticeable by the user at a given distance before actually outputting some stimulus? $\endgroup$ – Girauder Oct 10 '16 at 9:39
  • $\begingroup$ In order to compute d' you need to record an approximation of the guess rate too. This is often referred to as bias, for example, perceptual bias. Let's say you have two conditions: target present, target not present. Instead of giving the user a response option to press a button when the user sees the target, you tell the user to press one of two buttons: Target present, Target not-present. It follows that you will have four different response types: Hits, Misses, False alarms, and Correct rejections. (In memory research, a common task is the delayed match - hence "matching"...). cont. $\endgroup$ – noumenal Oct 10 '16 at 12:11
  • $\begingroup$ If you know the distance from the user to the screen you can compute the visual angle corresponding to the size of the stimulus, but if you can keep this distance constant, you don't have too compute individual values. // When comparing two different stimuli, you can present them in random order and compute d' individually for each condition (i.e. stimulus). You can then compare the ROC curves for both stimuli in order to determine the one that is most salient - at least for the specific viewing circumstances. However, you need to specify a criterion, for example: 75% correct or 95% correct. $\endgroup$ – noumenal Oct 10 '16 at 12:15

As with many such questions, this depends on your definition of (visual) salience. Your question suggests that you want to compare this across sensory modalities (auditory and visual) which comes with considerable difficulties.

One possible definition of "salience" is the property of a stimulus which makes it stand out from its background and grab our attention (see http://www.scholarpedia.org/article/Visual_salience). You could measure this empirically by asking observers to find/respond to the item in various different locations. They should be quicker to do so if this item is more salient (all other things being equal). This is essentially the logic behind most experiments into visual attention and visual search. As mentioned in another answer you could also monitor where the observer is looking, because salient things should be looked at earlier and more often.

There is now a huge amount of research into modelling image-based, visual salience. In theory, such models aim to take a picture of a scene and predict which parts of the image are going to attract attention. Another way to quantify the salience of different items would therefore be to measure the contrast or activity in the saliency map for particular locations. There is much more about this, and many references, in the scholarpedia reference above.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.