Abstract

F. Baluch, L. Itti, Learning in an attentionally-demanding triple-conjunction task, In: Proc. Vision Science Society Annual Meeting (VSS09), May 2009. (Cited by 34)

Abstract: Several studies have shown improvement in perceptual discrimination over the course of training sessions with spatial or feature range specificity. This type of specific learning does not address the question of whether and how general visual task expertise can be gained. We designed a conjunction search task in which a different target was previewed on each trial and was then randomly positioned in a search array, thus eliminating both feature and spatial specificity of learning. 5 subjects performed 10 sessions of 100 trials each over the course of 10 days. In each trial subjects had to find a color gabor patch target uniquely identified by its hue, orientation and spatial frequency among an array of 32 patches. All subjects showed marked improvement over the course of training. A highly significant (p<0.005) change in performance was observed from session 1 (44% correct) to session 5 (73% correct), beyond which performance plateaued. In order to isolate changes in visual behavior resulting from learning we construct feature similarity maps that define the similarity between items in a search array and a target, in individual feature dimensions. High values in the similarity maps indicate large differences between the target and an item, and vice-versa. We find small but negative correlations between saccade endpoint distributions (SED) and feature similarity maps, indicating that subjects were preferentially looking at distractor items more similar to the target. Moreover, the negative correlations grow stronger over the course of the sessions, for the hue and frequency features but not for orientation. We then correlate SED with linear combinations of individual feature maps as well as a map derived from a simple minimum distance rule. The results indicate that subjects had an increased tendency to look toward items similar to the target in the hue and frequency dimensions. Acknowledgement: NGA.

Themes: Model of Bottom-Up Saliency-Based Visual Attention, Model of Top-Down Attentional Modulation, Human Psychophysics, Human Eye-Tracking Research