I'm looking for a toolbox to model visual search performance in a singleton search task based on line orientations (you need to find a line that is most different from all others in it's orientation).

The desirable features would be:

  1. Feature-detectors that resemble real neurons.
  2. Memory for distribution of distractors in feature space during previous trials.

Is there something like that? I also know that sometimes similar tasks are studied in texture discrimination. Maybe there is something close to what I need in that domain?


I'm assuming that you want some kind of "computer vision" model (in that you want to be able to provide the model with input stimuli in the form of an image), and that you want to predict some kind of behaviour? (e.g., RT from a search task). Fleshing out the different processes involved is not going to be trivial, so there probably isn't a "one-size-fits-all" "toolbox" for your problem.

Perhaps the "saliency map model" by Itti and Koch would be suitable. There are many more modern versions of this (including a MATLAB toolbox and a number of related models). While it is not necessarily the best, the advantages in this case are that Itti and Koch actually used an orientation pop-out task in their original paper (Vision Research, 2000). However, I'm not sure to what degree the feature detectors actually resemble neurons, and I don't think a memory component has been included.

  • $\begingroup$ Thanks! The problem with the saliency models, as you correctly noted, is that they don't have memory. In addition, saliency is a measure of difference between a particular point and its neighbors (I am drastically simplifying, of course). So the distribution in feature space is not important - what's important is a spatial distribution of features and that's not what I need. $\endgroup$ – Andrey Chetverikov Jan 27 '16 at 18:35

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.