Abstract
Objective: Examine saliency of luminance/color, and test diagnostic/anti-diagnostic colors.
Methods: Used two sets of 8 512×512 images: diagnostic colors (example: red cherry); anti-diagnostic colors (yellow strawberry). Images contained 3×3 arrays of fruits/vegetables. Procedure within each set:
For each image, k=1–8, obtain (from HSV coordinates) the luminance image Lk (set H=S=0), and the color image Ck (set V to fixed value).
Form low- (l) and high-pass (h) versions (sigmas 38 and 10 pixels, respectively), resulting in 9 variants for each Fk: original Fk, Lk, Ck, and filtered versions, Flk, Fhk, Llk, Lhk, Clk, Chk.
Consider all possible pairs of images k and m, and form all composites Ikm with the color component of one variant and the luminance component of the other (2–9 such images for each pair).
Observers reported whether composite image Ikm resembled more Fk or Fm for all 684 ((9x8/2)x18) composites. We report on averages of 7 observers, who obtained very consistent patterns.
Results: 1) Lk dominated Cm, Clm and Chm at rates above 62%, 97%, and 95%. 2) Llk was dominated by Cm, Clm or Chm at rates above 90%. 3) Lhk dominated Clm and Chm at rates above 80%, but lost to Cm at above 62%. There were no major differences between diagnostic- and anti-diagnostic-color sets.
Conclusions: Generally, luminance is more salient than color. High-frequency components are dominant for both. Diagnostic colors were not advantageous over anti-diagnostic colors in our experimental design; experiments with single-item images and additional sigma values are in progress.