Be aware of noisy gradation

In short: Some existing ICC profile builders fail to make smooth gradation, especially for common printers. Zygomatic Color can do it.

In long: It is a long long story.

The pithole of some existing ICC profile builders

Let’s see the performance of some existing ICC profile builders.

The screenshot below shows the color of untuned gray gradation of each ICC profiles. Look at the red and blue curves. It represents a* and b* of CIELAB. Every curves are serrated.

Are these serrations real behavior? In a sense, yes this is real. When the printers printed the color targets, they fluctuated. The serrations are just noise and not reproducible. The serrations make noise on color conversion results, of course. You surely don’t want any noise like the serration. This is the pithole of some existing ICC profile builders.

So let’s look at ICC profiles built by Zygomatic Color.

How does Zygomatic Color avoid the pithole? By regression.

Interpolation

Interpolation fits curve (or polyline) to measured points. The curve is called spline. Spline is good for fitting, but not good for reflecting the behavior’s characteristic. Spline passes through all measured points.

Some existing ICC profile builders employ polyline. The design is simple, but susceptible to noise. The downside isn’t noticeable for expensive photo printers because they fluctuate little. Common printers fluctuate much. This is why some existing ICC profiles for common printers are always miserable quality.

Regression

Regression fits curve (or line) to measured points too. The curve isn’t spline (usually). The curve doesn’t pass through any measured points (usually). The curve is desinged from the behavior’s characteristic. The design of the curve is called model.

If the model doesn’t fit to the behavior’s characteristic, regression makes a disaster like below.

The behavior of tone reproduction may seem well known. In a sense, yes it is well known. The real problems are:

• The characteristic of our eyes.
• Computablity.
• Graceful error for the small gap between the model and the behavior.
• The behavior of printer drivers.

The model should have especially fine accuracy where our eyes are sensitive. The model should be computable (B2A in jargon). For these reasons, the model cannot be perfect. A small gap is inevitable. The gap shouldn’t make a disaster. Printer drivers… It is a sorcery.

How about the perfomance of the model of Zygomatic Color? Let’s see.

The accuracy of the regression model of Zygomatic Color

The histogram below shows the differences between model-predicted and measured colors of the patches in the printed and scanned targets.

The number above is a normal case. Your eyes usually cannot distinguish color difference under ΔE00 < 1.

When the gap between the model and the behavior is large, the model accuracy becomes bad. In case that the model doesn’t fit to the customer’s printing system, Zygomatic Color shows the model accuracy histogram while the customer is making their ICC profile.