1

As a conventional method of color camera calibration, I use following procedure

  1. Take a raw image of Xrite ColorChecker passport in a given light and exposure, and convert it to DNG file.
  2. Use Xrite software to create DNG camera profile using the image taken in Step 1.
  3. After loading the DNG camera profile in ACR, perform camera calibration after white balance.

I know something can go wrong in this procedure, but let's assume everything in the process was done accurately for now.

My question: what is the best process of validating this camera calibration?

Naturally, I tried to apply the profile to the image of Colorchecker passport, and examine XYZ values of patch area in the image with values calculated from reference values of corresponding color patch that Xrite provides with. But the difference between corrected and reference values seems to be large (> 20). I wonder if there is any better way to validate calibration result? Any advice will be appreciated.

  • Just to be sure : after the color calibration, have you corrected for white balance ? – Olivier Mar 23 '17 at 8:32
  • @Olivier I think so. I used White Balance Tool on color checker passport image. – Paul Mar 23 '17 at 9:51
1

Your process should give you the best possible custom white balance.

However, this may be all for naught if you are not using a recently calibrated and profiled monitor.

  • 3
    The monitor is irrelevant as the OP is measuring numerical values of the image used for calibration after having applied the calibration => the corrected numerical values should be the same as the ones claims by XRite – Olivier Mar 23 '17 at 8:32
  • So you are implying you validate calibrated image with your eyes by comparing checker passport and calibrated image on monitor, right? – Paul Mar 23 '17 at 9:58
1

I think I found an answer by extracting/analyzing DNG color profile. I used DCPTool to convert DNG profile to text file, and found that its content basically consists of three elements in case of single light source.

  • Estimated illumination: the light source estimation is not exact. It's just approximation to one of among 25 standard illumination sources such as D55, D65, Standard A, and so on.
  • Color correction matrix: 3x3 conversion matrix from camera color to XYZ (D50)
  • Information necessary to perform "hue twists": adjustment of hue and saturation of HSV based on value of HSV so that the twisted colors can be more recognizable by humans.

Based on these findings, my conclusion for possibility to validate the color correction using DNG profile with published CIE Lab values of color patches was very low if not none with following reasons.

  • Too rough estimation of illumination
  • No easy way to white-balance to D50 light source (e.g., with ACR/Lightroom)
  • Hue twists produce different resulting images from different exposure settings
  • If you want to capture and process an image of a colorchecker that will have the same measured colors you have to do something called "scene referred" processing. This is normally something done for reproduction of an existing image or art. – doug Mar 26 '17 at 5:38
  • @doug I think I know how to scientifically correct colors with a color target as there have been many methods suggested. I just hoped I could utilize common method with custom DNG profile for color calibration as it would save some time for programming. – Paul Mar 26 '17 at 23:07
  • 1
    This guy ludd.ltu.se/~torger/dcamprof.html#scene_referred has done some good work on creating a DNG camera profile for both output referred and scene referred use. If you want the best match to a colorchecker, use scene referred. His stuff is quite good. – doug Mar 27 '17 at 1:58

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.