Salience from feature contrast: additivity across dimensions

Test targets ('singletons') that displayed orientation, motion, luminance, or color contrast, or pairwise combinations of these, were presented in line texture arrays, and their saliences were quantified in comparison to reference targets at defined luminance levels. In general, saliency effects in different stimulus dimensions did add, but did not add linearly. That is, targets with feature contrast in two dimensions were generally more salient than targets with only one of these properties, but often less salient than predicted from the sum of the individual saliency components. Salience variations within a dimension were compared with and without a second saliency effect added. The resulting gain reduction in the combined stimulus conditions was interpreted to reflect the amount of overlap between the respective saliency mechanisms. Combinations of orientation and color contrast produced the strongest gain reduction (about 90% for color in orientation) thus indicating the strongest overlap of underlying saliency mechanisms. Combinations of orientation and motion contrast revealed about 50% overlap, slightly smaller rates were found for combinations of color and motion. All combinations with luminance contrast (orientation and luminance, motion and luminance) produced only little gain reduction (<30%) thus indicating a higher degree of independence between the underlying saliency mechanisms than for other stimulus dimensions.