This is a question I often get. Are errors in hue, saturation, or luminance most detrimental to an image? Fortunately, this question can be answered in a relatively straightforward way.
First, we have a very good metric for grading perceived color errorDelta E. There are several dE formulas, but for color I think that CIE94 is the best.
Second, we have clearly defined gamuts that we can use as a reference. For the purpose of this test, I'll use Rec. 709, the high-definition standard.
Third, measuring the amount of error is fairly straightforward as well. Color is a three-dimensional property, including hue, saturation, and lightness. We can assign numerical values to any color using u'v' or ab chromaticity coordinates for saturation and hue and the CIE L value for lightness.
Terms and Methodology
Here's the data.
10% HSL errors and CIE94
Conclusions
First, we have a very good metric for grading perceived color errorDelta E. There are several dE formulas, but for color I think that CIE94 is the best.
Second, we have clearly defined gamuts that we can use as a reference. For the purpose of this test, I'll use Rec. 709, the high-definition standard.
Third, measuring the amount of error is fairly straightforward as well. Color is a three-dimensional property, including hue, saturation, and lightness. We can assign numerical values to any color using u'v' or ab chromaticity coordinates for saturation and hue and the CIE L value for lightness.
Terms and Methodology
- Selecting an arbitrary amount of error, let's say 10%, then it is a simple matter to determine how much dE results from a uniform 10% error in each of the primary colors. All you need to know is how to identify quantities of error in HSL.
- Saturation is defined as the chromatic element of color minus lightness. Chroma is the square root of the sum of color coordinates (either u'v' or ab) each squared. You can also think of saturation within a defined gamut as distance from the white point. However, since the available chromaticity charts on which we would plot such distance are not perceptually uniform, this is an approximation only.
- Hue is defined as the degrees around a 360 degree circle with the white point at the center. Within a defined gamut, hue is the angle of a color from the white point. It is also arguably the most important characteristic, because it is hue that distinguishes one color from another.
- Lightness is a perceptually-weighted measurement of luminance. It is defined as 116 times the root cube of relative luminance minus 16. Relative luminance is just the measured luminance of a color as a percentage of white at the same level of stimulus.
- Although the calculations for HSL and dE require using the 1976 color spaces (Luv or Lab), values will be presented here in xyY simply because this is more familiar.
Here's the data.
10% HSL errors and CIE94
| Red | ||||
| x | y | Y | CIE94 | |
| H | 0.603 | 0.292 | 0.2126 | 11.8 |
| S | 0.669 | 0.330 | 0.2126 | 6.3 |
| L | 0.640 | 0.330 | 0.2658 | 5.5 |
| Green | ||||
| H | 0.272 | 0.580 | 0.7152 | 5.8 |
| S | 0.298 | 0.644 | 0.7152 | 3.7 |
| L | 0.300 | 0.600 | 0.9120 | 8.9 |
| Blue | ||||
| H | 0.175 | 0.059 | 0.0722 | 4.7 |
| S | 0.142 | 0.046 | 0.0722 | 4.3 |
| L | 0.150 | 0.060 | 0.0879 | 3.5 |
Conclusions
- From this it seems clear that hue errors are the most perceptually noticeable. A 10% error in hue results in the highest average dE of 7.4 and is the most noticeable for blue and red and the second most noticeable for green.
- Determining the second most noticeable type of color error is less clear. Although lightness errors result in a higher average dE than saturation errors (6.0 vs. 4.8), saturation errors are more noticeable for red and blue where lightness errors are the least offensive. Only with green is a lightness error more noticeable (by a relatively large margin) than an equivalent saturation error.
