Accurate Image Manipulation for Desktop Publishing

Profiles or Pro Files?

In the quest of affordable ICC profiling approach for system calibration I have detected several quality issues. This page explains two major findings, inaccuracy of the profilers and clipping of the data. (inaccuracy of proprietary Q-60 -like targets and scan-to-scan repeatability of scanners are currently under preparation).

The performance of:

Rather expensive MonacoEZcolor profiler (300 USD).
IPhotoMinusICC freeware profiler
CGI Assisted Manual Calibration

are compared in scanner calibration. Profilers that cost several thousands of US dollars will do better than the two above, but usually such an investment can not be tolerated. We really should get decent accuracy built-in into our digital imaging equipment's.

General

All the work was conducted in higher-bit-depth in WideGamut, D6500 gamma 1.8 working space.
Scanner was Umax PowerlLook III.
Target was Kodak Q-60 Reflective R1199704.
Images shown on this page are copies of the final images that are published (profile converted) from the WideGamut, D6500 gamma 1.8 working space into the nativePC profile, therefore this page has to be viewed using an uncalibrated PC system (or the gamma control of the system has to be set to 2.5).

Workflow

Scanner was warming up about 30 minutes with the lamp ON. Twain driver was set to full manual, gamma 1.8 and levels to 0 ... 255. The Q-60 target was scanned.
The Q-60 scan was used in creating the scanner ICC profile with MonacoEZcolor and with IPhotoMinusICC and the CGI Assisted Manual Calibration was created with it also. Same source file for all the three calibrations.
The scan was then profile converted using the MonacoEZcolor profile and IPhotoMinusICC profile and using the CGI Assisted Manual Calibration.
Copies of the results were puplished to nativePC profile (Triniton, D6500 and gamma 2.5) and uploaded.
Clipping was inspected using a grayscale CGI, it was again converted using hte both profile and using the CGI Assisted Manual Calibration.

Is the Q-60 CGI simulation accurate?

In case you doubt the accuracy of the Q-60 CGI simulation please download the Q-60 R11997:04 CGI simulation and its datafile from the Kodak's site. Then:

Open the Q-60 simulation into Photoshop, allow Photoshop to convert it into your working space.
Open the Kodak datafile into a text editor such as the Notepad.
Set the eye-dropper tool or the color-sampler tool output in the Info-palette to Lab mode
Start comparing the Lab values against those in the Kodak datafile, patch by patch.

The above is a very easy way to verify the accuracy of the Q-60 CGI Simulation. Do it if in doubt, it is interesting experiment, however the Q-60 CGI simulations are very accurate.

Calibration Comparison

Slowly toggle between the Q-60 CGI Simulation and one of the calibrated results to see difference.

The IPhotoMinusICC profile is slightly towards blue and more notable it is scaling the image up, making it more light. For the scan of Q-60 this does not do harm because the Q-60 target completely lacks highlight info. But it will have strong adverse effect in the actual work, as is explained below in the clipping comparison -section.

Monaco EZcolor is also scaling the image data up slightly, and in addition it applies a slight up-curve resulting rather inaccurate calibration.

Both the MonacoEZcolor and IPhotoMinusICC converted versions appear to have more "punch" but it is only because the CGI Assisted Manual Calibration does not clip the highlights.

Clipping Comparison

In order to experiment how the profiles act over the full intensity range a simple CGI that has a patch for every 256 gray values was converted using the three methods.

Is it meaningful or practical to test an ICC profile using such a 256 patch grayscale CGI, because it is not coming from the scanner?

Well, if the range of the scanner is from level 0 to level 255 (like it always is) then the 256 grays will show if the profile is clipping the data that the scanner outputs.

The clipping in the black end is shown by a white border and clipping in the white end using black border.

As said above in the general section also these conversions were performed in the 16-bit per channel mode so there is no quantization issues due to 8-bit manipulation.

There appears to be severe clipping problem with the IPhotoMinusICC in the highlights and a hue-error just before clipping. It also clips strongly at the black end.

MonacoEZcolor has small clipping problem in the white end and it too clips strongly at black-end. It also wants to think that the scanner white is neutral white even if grays are not.

CGI Assisted Manual Calibration will clip that amount that it is designed to clip.

Note: When working in a steep gamma space like the 1.8 there will always be some clipping in the black-end since several levels there are totally wasted due to the extreme non-linearity of the gamma function.

To avoid such clipping in the white-end one could perhaps consider adjusting the output range already in the scanner driver... but it will not help at all with the situation. One can either:

Clip the range already in the scanner driver a little... but this only changes the place where the clipping occurs.
Scale the image data down a little in the scanner driver... but then the profiler will do what the profiler will do, it still will clip.

Note: Please do not download images that are shown on this page and draw conclusions based on them. They do show the results correctly but for reconstruction of thees experiments they are not accurate enough because they are copies 0published to the nativePC space for viewing with uncalibrated PC. Instead download the icc_experiment_01 zip archive (763kB) it contains all the required material needed for reconstruction.

Accurate Image Manipulation for Desktop Publishing