Summary of the methodology
- Stimuli displayed on a BARCO monitor connected to a DELL (Precision 6.1.0) PC (the resolution was 1280×1024, 256 colours, and the frame rate was 75Hz).
- The stimuli consisted of a bipartite field (10×10 deg.) partially overlaid by a simulated square (5×5 deg.), displayed at the center of the monitor.
- A total of 240 stimuli were selected: 60 per each chromatic change, 20 per each luminance level and 2 per each vector length. Four types of color change were tested: pure translation, pure convergence , shear, and divergence in the CIE XYZ space.
- Three conditions accounted for luminance: vectors in the equiluminant plane, vectors that point to a lower luminance (filter) and vectors directed to a higher luminance (illumination). Ten different vector lengths were assigned to these conditions.
- Six observers were selected for their normal or corrected-to-normal visual acuity and normal color vision, according to the Farnsworth-Munsell 100-Hue-Test. The set of all patches was presented in a randomized sequence. For each patch, the subjects responded (with pre-established keyboard commands) whether they saw a transparent overlay or not, or whether they did not know. Each stimulus was shown for two seconds.
- A Log-Linear Model was used to calculate the relationship between the different variables. We evaluated the model fitted to the data with the Pearson chi-square statistics.
Example of the stimulus configuration.
Summary of the results
- According to our analysis, convergences, despite luminance and vector length variations, appear transparent. However, large equiluminant translations do not lead to the perception of transparency. Also, small shears and divergences produce transparency. These results are inconsistent with D’Zmura et al’s Convergence Model of transparency perception.
- First, it appears that translations are not equal across luminance level variations: we have shown that subjects have more difficulties seeing transparency based on equiluminant translations than on equiluminant convergences. Our data provide that observers do not seem to see transparency for all equiluminant translations. Longer length vectors show less transparency. Luminance is one of necessary parameters for translation, and vector length also. When we present equiluminant translations to the subject, they appear transparent depending on vector length. We have found that convergence does not give such a result, probably because all our vectors are directed to the same hue. This might create a sufficient perceptual link between the overlay and the original surface. However, it was found by Chen et al. that the Convergence Model does not lead to the perception of transparency when surface colors have opposite hues or lightness. We assume that, for translations, when the vector based at a given hue exceeds a certain locus around this hue (e. g. directed to an opposite hue), the chromatic contrast is too large to give the impression of transparency.
- Second, we wanted to test whether other chromatic changes, such as shear and divergence, do not give the perception of transparency. Nevertheless, our results also show that for short vector lengths, observers do detect transparency with shear and divergence. We can suppose X-junctions are sufficient cues to perceive transparency under these conditions, and that for small chromatic changes, this is enough to generate perceptual scission, despite any particular chromatic change. Small shears or divergences tested using an L-junction configuration could evaluate this hypothesis.
Results for convergences.
Results for translations.
Results for shears.
Results for divergences.
Swiss National Science Foundation (SNF) under grant number 20-59038.99.