PURPOSE:  A salient feature contrast between a target item and many background items results in perceptual pop-out of the target from the background. However, when the feature contrast is small the target may not be salient initially, but sufficient training on the detection task can also result in a pop-out e?ect. Previous studies suggest that the pop-out effect induced by simple feature contrast is mediated by primary visual cortex (V1) through creating a saliency map, which is thought to be a bottom-up process. In the current study we examine whether the saliency map in V1 can be enhanced by perceptual learning.
METHODS: Two monkeys were trained to detect a line segment whose orientation was deviated by varying extents from a background of iso-oriented lines. Over the course of training we recorded V1 neuronal responses with implanted microelectrode arrays. We used support-vector-machine (SVM) classifier to decode V1 responses.
RESULTS: The animals' performance in the detection task, measured as percent correct and reaction time, was signi?cantly improved with training for targets with small orientation contrasts. By comparing neurons with different preferred orientations relative to the target, we dissociated two distinct neural response components that signal the pop-out e?ect. One component is stimulus-driven and little affected by training; it relies on orientation-dependent contextual modulation mechanisms, which create a saliency map from neural response outset. The other component is mainly task- and experience-dependent; it signals the perceptual saliency of the target, independent of the preferred orientations of neurons. The latter mechanism generates a delayed saliency map (40 ms after visual response onset), which is enhanced by perceptual training and correlated with improved behavioral performance.
CONCLUSIONS: V1 is involved in detection of local-feature contrast through two separate saliency maps. Perceptual learning enhances the late saliency map but not the early, bottom-up, saliency map.