THE COLORCHECKER PAGES (2/3)

Click to go to a section of interest:

Page-1: Attention ! Warning ! Important ! Read-me ! A brief presentation ColorChecker charts formats / pictures New color specifications since November 2014 Where to buy? Page-2 (this page): ColorChecker data (RGB coordinates, spectral data, Excel tables) Visual comparisons (BEFORE Nov. 2014 vs AFTER Nov. 2014) Visual comparisons (individual charts vs average) ColorChecker images (RGB and L*a*b*; 8/16-bit) Do you want to provide some data? Page-3: Using the chart to auto-correct images or make ICC/DNG profiles Using the chart to analyze your camera's performance The problem with the ColorChecker SG

ColorChecker data

The data presented here is valid for the following charts (see descriptions and images of these charts): • Large size: ColorChecker, ColorChecker Classic • Mid-size: ColorChecker Proof • Small-size: ColorChecker Passport, ColorChecker Classic Mini, Mini ColorChecker It is important to mention that we see no significant statistical differences in the colors of the patches measured on two ColorChecker Passport and two ColorChecker Proof relative to the average previously measured with a mix of 20 ColorChecker and Mini ColorChecker. This is why the data from ColorChecker Passport and ColorChecker Proof targets was incorporated in the average of 30 charts. However, the data presented here is not valid for the ColorChecker Digital SG. Measurement geometry: 45 deg./0 deg. for all data sets. The illumination is an inverted light cone hitting the color patch at 45 degrees and the measurement is taken perpendicularly to the patch, at zero degree. This configuration is such that the specular (i.e. mirror-like) reflection from the light on the patch surface is not directed back on the sensor, and is thus less sensitive to gloss; this is the preferred geometry for printing applications. Please note that this geometry is considered the same as when the Illumination hits the target perpendicularly and the measurement is taken at 45 degrees (i.e. 0 deg./45 deg.). Measuring instruments: 24 of the 30 data sets were measured with Eye-One Pro (i.e. i1Pro) spectrocolorimeters from X-Rite/GretagMacbeth, three with the X-Rite DTP20 (Pulse), two with the GretagMacbeth Spectrolino/Spectroscan, and one with the X-Rite DTP22 (Digital Swatchbook). The DTP20, DTP22, and Spectrolino/Spectroscan are now discontinued. The 24 i1Pro used in this compilation were a mix of Rev-A to Rev-D instruments, manufactured between 2003 and 2011, with most of them non-UV cut (although this has no effect on the measurements since the ColorChecker pigments are not fluorescent). None of the measurements were derived with the X-Rite XRGA protocol, a protocol designed to improve inter- instrument agreement between many legacy and current X-Rite and GretagMacbeth instruments; announced in 2010, XRGA is progressively being integrated in X-Rite products. As per X-Rite XRGA Web page, the changes induced by the XRGA protocol should be minimal. In April 2012, X-Rite has announced the availability of the Eye- One Pro 2 (i1Pro 2), which replaces the original Eye-One. Based on the same measurement geometry and as the model it replaces, the i1Pro 2 has new and improved features, one of them being the capability to support the M0/M1/M2 Measurement Conditions as defined in ISO 13655. When making measurements of the ColorChecker with the i1Pro 2, the M0 condition should be selected as it is equivalent to what we obtained with the older Eye- One, and is still perfectly adequate for non-fluorescent substrate measurements. The first document is a tutorial-type article which presents the RGB coordinates of the ColorChecker patches in 4 common RGB spaces: Apple, Adobe (1998), ProPhoto and sRGB (Note: Additional data for these and 15 other RGB spaces can be found in the spreadsheet below). The coordinates were derived from the GretagMacbeth (2005) L*a*b* D50 reference data and from the averaged spectral measurements (of 20 charts); the two data sets are compared against one another and against other references, such as the 1976 McCamy article. For those who want to dig deeper, this document also describes the data conversion procedures and presents all the required equations. • RGB coordinates of the Macbeth ColorChecker (319 KB) (June 2006) A PDF document on the RGB coordinates of the ColorChecker in 4 RGB spaces (Apple, Adobe, ProPhoto, and sRGB), and the method by which they were derived. The second document contains numerical data and graphs. The average data is derived from measurements of 30 charts (formulations BEFORE November 2014). For those who have downloaded the previous spreadsheet (issued in 2006) which contained average data from 20 charts, you will notice that the average data from 30 charts shows minor changes and no surprises; the changes affect about half of the patches, with one or two RGB component changed by 1 unit (in 8 bit). However, even if these changes are minor, we have updated the images from the averaged data to match the newly computed values. • ColorChecker_RGB_and_spectra.xls (155 KB .zip) (April 2012) This Excel spreadsheet contains: o RGB tables for 19 spaces, and L*a*b* (D50) values derived from averaged spectrums; o RGB tables for 19 spaces, and xyY (D50) values derived from GretagMacbeth L*a*b* D50 data (2005); o RGB tables for 19 spaces, and L*a*b* (D50) values derived from CIE (1931) xyY data (1976); o averaged spectral data with Standard Deviation tables (from 30 charts); o graphs of all the average spectrums with standard deviation. Note: eciRGB_v2 data is now computed in the April 2012 spreadsheet instead of eciRGB Version-1 data. For those interested in comparisons, here is the older spreadsheet for 20 charts: ColorChecker_RGB_and_spectra-Avg20_2006.xls The following documents contain ColorChecker data in industry standard CGATS, CxF2, and CxF3 file formats. • CC_Avg30_spectrum_CGATS.txt (11 KB) (April 2012) The ColorChecker averaged spectral data from 30 charts in CGATS format (a human readable text file). You can view the patches on your screen by opening the file with PatchTool (even in demo mode). For those interested in comparisons, here is the older file for 20 charts: CC_Avg20_spectrum_CGATS.txt • ColorChecker_CxF3.cxf (16 KB) (2012) Note: This file is zipped (*.zip) so that it does not open in the browser. The ColorChecker averaged spectral data (from 30 charts) and the X- Rite/GretagMacbeth L*a*b* D50 data (2005) in CxF Version-3 file format (Note: The CxF3 file format is used, for example, when saving measurements taken with the i1Profiler software from X-Rite). The file also contains the averaged data in L*a*b*, Adobe RGB, and sRGB coordinates as well as sRGB and Adobe RGB data derived from the X-Rite reference. Using PatchTool in demo mode, you can import spectral data or L*a*b* data plus one RGB space. • ColorChecker_CxF2.cxf (16 KB) (April 2009) Note: This file is zipped (*.zip) so that it does not open in the browser. The ColorChecker averaged spectral data (from 20 charts) and the X- Rite/GretagMacbeth L*a*b* D50 data (2005) in CxF Version-2 file format (Note: The CxF2 file format is used, for example, when saving measurements taken with the X-Rite ColorMunki). The file also contains the averaged data in L*a*b*, Adobe RGB, and sRGB coordinates. Using PatchTool in demo mode, you can import any of the saved formats, including both the averaged measurements and the X-Rite data in a single Compare window (select the "Quality-Control data" type in the "Import CxF2" dialog when you open the file). • ColorChecker24_Before_Nov2014.txt (948 bytes) (2005) The reference L*a*b* data from X-Rite for charts manufactured BEFORE November 2014. This is the data used for comparison with our average of 30 charts; the data was first made available in 2005. • ColorChecker24_After_Nov2014.txt (946 bytes) (October 2015) (New file - 2016/01) The reference L*a*b* data from X-Rite for charts manufactured AFTER November 2014 (including November 2014 according to a comment in the file!); this data is available since October 2015. Why was this data compiled? Most ColorChecker owners will have a need for the actual RGB values of the patches at one time or another. Until late 2005 this information was hard to find since the RGB values provided with the chart corresponded to no known RGB space. The only reliable reference data was xyY coordinates determined with Illuminant-C, as they appeared in the 1976 McCamy article. It was, and still is, possible to find, on the public Web and in scientific sources, published data obtained by people who measured their chart. While OK in principle, this is not enough (for me at least). How old was the chart? Is the chart representative of the production average? How precise was the measuring instrument? Was the instrument calibrated? How was the data processed afterwards? Should I say more? These and other factors make the results of a single measurement unreliable. This was the justification to write the first version of the "RGB coordinates of the Macbeth ColorChecker" pdf. This was also the reason behind the project to average spectral data provided by chart owners. Why is spectral data important? Because spectral information gives you the capability to determine the color coordinates as seen under any illuminant, and thus any RGB space, with more precision. Analysis of the spectral data also gives some hints that various pigments are/were/could-have-been used to manufacture the charts. This figure shows two types of spectrums seen for the light-skin patch; these two types are seen with both the Standard and Mini charts, with either new or somewhat older (3 years +) charts. Other patches show a similar behavior. It is important to note that these different spectrums do not necessarily result in distinctly different colors (or color coordinates) and other factors contribute to make the spectral data more fuzzy, such as storage effects, frequency of use/light exposure, and chart edition (manufacturing variability). One of the nice surprises in this analysis is that the charts seem to age very slowly when kept in dark storage and used on an occasional basis, which is the case for most serious amateurs and pros. A 12 years old chart (well kept, I insist) has 23 patches out of 24 within the standard deviation! One of the patches that varies the most is the White one, as can be seen in the Standard deviation table of the spreadsheet. I have recently seen a white patch, from a never-used ColorChecker Proof (i1Publish), which is quite yellow (L*a*b* D50, average of 3 measurements: 96,48/- 0,53/5,84) even if it was always kept in its black paper sleeve; however, it is impossible to say if this yellowing is due to the sleeve or if this is a statistical outlier, since other ColorChecker Proof were close to the computed average. In any case, the white patch is not the most neutral of the gray ColorChecker patches and the third neutral patch from the left is often recommended to perform gray balance. When this project began, we could compare only the data derived from the 1976 McCamy article to a handful of average measurements. Because experimental measurements with a few samples cannot be considered reliable, I placed, at the time, more importance on the data derived from the article than on the measured data. As the number of charts measured grew, it became obvious that the measured data was getting stable but still different from the original reference values. While the accumulating evidence was that user measurements were a more accurate assessment of the "average" chart than the data in the 1976 article, RGB coordinates and images (synthetic charts) determined using both approaches were made available on this site so that users could select the ones they preferred. Around October 2005, GretagMacbeth updated its reference data on the ColorChecker and made it publicly available (ColorChecker reference pdf) on the X-Rite Web site (link N.A. in 2022). This data includes L*a*b* D50 and sRGB values. However, they did not include RGB values for Adobe(1998), ProPhoto or any other RGB space. This 2005 GretagMacbeth L*a*b* D50 data matches the average measurements very well. On the other hand, both the average measurements and the new GretagMacbeth data do not match that well what was derived from the xyY coordinates of the 1976 article. These comparisons are discussed further in the "RGB coordinates of the Macbeth ColorChecker" pdf. As a result, the 1976 data can be considered obsolete; nonetheless, for completeness, the coordinates derived from the 1976 data can still be found in the Excel spreadsheet presented above. In the "RGB coordinates of the Macbeth ColorChecker" article, the GretagMacbeth L*a*b* D50 data is used to derive the RGB coordinates for Adobe(1998), Apple, ProPhoto, and sRGB (even if coordinates for this last space are proposed by GretagMacbeth/X-Rite). Of course, the RGB coordinates obtained from the averaged measurements are also provided for comparison purposes. In addition, the RGB data is presented in both 8-bit and 16-bit format, the later in full precision, not just a multiple of the 8-bit format, i.e. not just 8-bit values x 256. Finally, while the match between the RGB coordinates derived from the GretagMacbeth L*a*b* D50 data and the RGB coordinates determined from the average spectral measurements is quite good, the match between the sRGB values given by GretagMacbeth and the sRGB coordinates determined from average measurements is not as good. The results show it is likely that the GretagMacbeth sRGB data was derived from another data set than their L*a*b* data; much more details on this analysis are given in the pdf. I thus recommend the sRGB coordinates derived from GretagMacbeth L*a*b* D50 to the sRGB coordinates given by GretagMacbeth. So, with all this data and analysis, could we say that we have sufficiently characterized the ColorChecker? Almost! However, in October 2015, X-Rite announced that new color formulations have been used since November 2014, and that new colorimetric references were required. Thus, there is again a need for synthetic images which precisely represent the new formulations; this is why you will find images based on the new file reference in the ColorChecker images section. Of course, we will also need user measurements to validate the new references. Recommended data As for which data I think is better, the data from the average measurements or the GretagMacbeth/X-Rite values, here are suggestions: • If you plan to use ONLY YOUR ColorChecker chart, then use the actual measurements of your chart (of course if you have data for the specific RGB space you work in). • If you are looking at the image of an unknown chart, or if you want to print a "synthetic" chart that will be compared against an unknown "real" chart: o For charts manufactured BEFORE November 2014, use the data and images derived from our average measurements. o For charts manufactured AFTER (or during) November 2014, use the data and images derived from X-Rite data. • For charts manufactured BEFORE November 2014: o For L*a*b* D50 data and for D50 RGB spaces, such as ProPhoto, I consider that the values deduced from our average measurements and the ones from/derived from GretagMacbeth data are equivalent. o For D65 RGB spaces, I see a slight advantage in values deduced from our average measurements, because of the imprecision brought-in by the use of the chromatic adaptation transform (Bradford matrix) when transforming the GretagMacbeth data; such an imprecision is not present in our measurements since we have complete spectral information. o If you still prefer to use GretagMacbeth/X-Rite data for sRGB, I suggest using the sRGB data I derived from the L*a*b* D50 data instead of the sRGB values they provide.

Visual comparisons

(Before Nov. 2014 vs After Nov. 2014)

The charts manufactured BEFORE November 2014 and those manufactured AFTER November 2014 have different color formulations. The average of chart measurements presented in these pages applies to charts manufactured BEFORE November 2014. This section presents visual comparisons of the BEFORE and AFTER reference files, from X-Rite, and comparisons of our average data with both files. In the screenshots below, the top screenshots show the patch colors (an sRGB monitor is assumed) while the bottom screenshots show the color difference statistics using the CIEDE2000 color difference formula. Patches with a difference higher than 1.3 are flagged with a red border while patches with a difference higher than 1.0 but less than 1.3 are shown with a yellow border. The two screenshots on the left show a comparison between the BEFORE and AFTER reference files. The two middle screenshots show a comparison of our measured average against the AFTER November 2014 reference, and the two screenshots on the right show a comparison of our measured average against the BEFORE November 2014 reference. Note: In these screenshots, each square is made of two triangles such as this one: , with the top triangle assigned to the first color list being compared and the bottom triangle assigned to the second color list. Differences between the old and new formulations (BEFORE vs AFTER) are visible for many patches. In particular, the white patch is not as white (L*=95.2 instead of 96.5), and yellower (b*=2.9 instead of 1.2). The average color difference is 0.85 and the maximum difference is 1.95, with 7 patches with a difference larger than 1.0. A comparison of our measured average with the new formulations reference file (the two middle screenshots) shows very similar results, both in terms of which patches exhibit more error and in terms of the average color difference. A comparison of our measured average with the old formulations reference file (the two screenshots on the right) shows a very good match, with an average color difference of 0.49 and a maximum difference of 1.1. We see that the old X-Rite reference file and our measured average (of charts manufactured BEFORE November 2014) show very similar differences when compared against the new X-Rite reference file. This can be seen as an indirect validation of our measurement data, which is confirmed by the much better match seen between our average measurements and the old reference file. These comparisons confirm the need of a new reference file for ColorChecker charts manufactured AFTER November 2014. Since we do not have sufficient measurements to provide a statistically valid average for the new formulations, we recommend using the new X- Rite reference data (AFTER November 2014) for the time being; this file is also available in the ColorChecker data section on this page. If you are interested in providing measured data from your ColorChecker chart manufactured AFTER November 2014, please read this section.

Visual comparisons

(individual charts vs average)

The three PatchTool screenshots just below are examples of individual charts compared to the average of 20 charts (formulations BEFORE November 2014). The first example is from the chart whose average error is the smallest relative to the average chart, i.e. the "best". The second screenshot compares a typical "middle of the pack" chart to the average, i.e. the "median". The third one compares the chart which is the most different from the average, i.e. the "worst". Note: In these screenshots, each square is made of two triangles such as this one: , with the top triangle assigned to the average. For many patches, especially the ones of the "best" and the "median" charts, the triangles cannot be distinguished since the difference is very small; enlarging the chart so that it fills the screen will help, as well as taking the time to look for differences (5 to 10 seconds for each patch). To help classifying the patches, colored borders are added when color difference thresholds are exceeded. These borders are yellow and red in the above examples, with the yellow border applied to all patches whose difference is larger than the value assigned to the lower threshold but smaller or equal to the higher threshold, and the red border applied when the color difference exceeds the higher threshold. In practice, 50% of the charts should be similar to, or better than, the median chart, with essentially one or two color differences which are more easily noticeable. The next three screenshots respectively show the standard deviation in Lightness (L*), Chroma (C*), and hue (h) computed from the 20 charts. The top triangle shows the average value minus the standard deviation, and the bottom triangle shows the average value plus the standard deviation. For a given screenshot, only one parameter is different, either L*, C*, or h, and the other two are the same. Expressed in terms of statistics, there is 68% probability that your chart patch is between the top and bottom triangle for this parameter. As we see in the screenshots, the Lightness variation is the most important factor, followed by Chroma, then hue. The Chroma and hue differences are particularly small and difficult to notice. In the third screenshot, all neutral patches are assigned a red border, corresponding to an apparently large standard deviation in hue higher than 20, even though the difference is not visually noticeable (Deltah is large but DeltaH* is small); this is expected for very neutral tones, near the illuminant, for which the Chroma (saturation) is very low. Finally, to get an idea of the maximum color difference that you could encounter between your ColorChecker and the average values, we used the CIEDE2000 color difference formula and, for each patch, compared the average value (average file data) to the individual data of the 20 charts. We then selected the "worst case" for each patch, and built a composite color list ("worst case" file data); this list is made of patches from 12 of the 20 charts (i.e. all the patches of the other 8 charts are better than these worst cases). The top triangle is the "BabelColor Average" patch, and the bottom triangle is the "worst case" patch. Please note that the "worst case" patches are NOT representative of the statistical "standard deviation", which is much less, as can be seen in the three screenshots just above. Worst patches of 20 charts You will (hopefully!) never obtain a comparison as in this worst case screenshot when comparing your chart to the average, since a typical chart may contain between zero and four similar patches, and finding between 0 and 2 is more likely. Also, you will note that six patches have a CIEDE2000 color difference of less than one, which is excellent, especially considering these are the "worst cases"!

ColorChecker images

Click on one of the following ColorChecker icons to see the image, or right-click to download the linked file or the target. The QuickTime plug-in may be required to view TIF images in a browser, which will not show the colors correctly anyway since these images have NO embedded profiles. In addition, 16-bit TIF files are not well supported in Web browsers. It is thus strongly recommended to first download these images and open them in a graphics editing program afterwards. Simulation images of the ColorChecker RGB-TIF 8/16-bit, BMP/PNG 8-bit, and L*a*b* 8/16-bit (complete sets and some files are zipped) The images "From GretagMacbeth L*a*b* D50" were derived from the reference data published by GretagMacbeth in 2005. The L*a*b* data was converted to D65 with the Bradford Chromatic adaptation matrix when required, and then converted to RGB. Because, as we mentioned, these images have NO embedded profiles, each of them should be assigned its respective RGB space profile (the Adobe RGB profile to the Adobe file, etc.) using Photoshop or any other program which can assign a color profile to an image. They are referred to as “formulations BEFORE November 2014” since X- Rite announced that new color formulations were used for chart production starting at this time. Data tables for these images and fifteen other RGB spaces are also available in an Excel spreadsheet. The images "From the average of 30 charts" were derived by averaging spectral measurements sent to me by YOU! The data was measured from various ColorChecker chart formats presented in Page-1 of the ColorChecker pages, except the ColorChecker Digital SG. All charts used for this average correspond to the formulations used BEFORE November 2014. Data tables for these images and fifteen other RGB spaces can also be found in an Excel spreadsheet. Please note that we currently do not have enough measurement data to compute a valid statistical average of charts manufactured with the new formulations (since November 2014), and so it is useless to create images based on measurements for the moment. The images "From X-Rite L*a*b* D50" were derived from the reference data made available by X-Rite in October 2015. The data is to be used in conjunction with charts manufactured AFTER (and during) November 2014. The L*a*b* D50 data was converted to the RGB spaces with the same method used for the previous reference data. You will notice that the bitmap (.bmp) images were replaced by PNG images and that we do not provide images in the Apple RGB space anymore. Data tables for these images should be available in the future. Note for 16 bit images: The 16 bit RGB and L*a*b* images were obtained programmatically from the spreadsheet data values since it is not possible to assign 16 bit colors in most image processing software. While these are full 16 bit images, the colors in these images will only be quantified in 15 bit+ by Photoshop; please consult the PatchTool Help manual for more information on the 15 bit+ vs 16 bit formats. A 16 bit image can easily be generated by PatchTool; you first need to import a 16 bit RGB color list, a L*a*b* color list, or a file with spectral data such as the "CC_Avg30_spectrum_CGATS.txt" file available above, then export as an image in 16 bit TIF format. Note for L*a*b* images: The 8 bit L*a*b* TIF images may contain color values slightly different from those in the spreadsheet; these differences are sometimes visually noticeable. This is due to how L*a*b* values are encoded in these files; in particular, a* and b* can only be assigned in integer increments. Not decided on which image is better? Read this. These images and spreadsheet could be updated as I get more data. Thanks in advance! If you are not satisfied by the looks of the chart image on your display, you are not the only one! This is not a new problem, as you can read in this 2004 article, Colour & Computers, by Charles Maurer, in Tidbits, an online publication for the Macintosh Internet community which is in its 25th year! His article has a few paragraphs on his experience of the various renderings of the ColorChecker chart obtained with different flavors of the sRGB profile and different applications. Monitor technology and software have changed, but not our requirements. Nowadays, we can work with wide-gamut displays with built-in Look-Up-Tables bundled with colorimeters and custom calibration software, from NEC and Eizo for instance, which are close to perfect. Even an entry-level display calibrated (i.e. profiled) with an entry-level colorimeter will provide a tremendous improvement in color accuracy and image editing efficiency.

Do you want to provide some data?

At the moment, we are particularly interested in obtaining measurements from charts manufactured since November 2014. You can contribute to this study if you have a ColorChecker chart, and if you can generate a spectral data file from a spectrophotometer. I will compile the results and make them available on this page. If you have a colorimeter with which you can measure only the color coordinates, without spectral data, well, send your data also! Preferred instrument and measurement conditions: • i1Pro, i1Pro 2, or any instrument with 0 deg./45 deg. geometry • Measurement Conditions: M0, i.e. tungsten lamp, i.e. Illuminant A (the effect of the other conditions should be small for this chart) • If possible, do 2, 3 or more measurements and average them (send all files). Measurement software and procedures: • i1Profiler: Use the "Measure Chart" or "Measure Reference" procedures in the Workflow Selector. • PatchTool: Use the Patch-Reader tool. You can do the measurement with a custom-defined chart or by using an opened ColorChecker file as a guide. • Eye-One Share (i1Share): Please read this procedure. • Measure Tool / ProfileMaker: Please read this procedure. Suggested file formats (CGATS file example): • CGATS: This is an ASCII text file format, often with a *.txt extension, supported by most color-measuring software, including i1Profiler, Measure Tool (part of ProfileMaker), and PatchTool. • CxF (*.cxf): Version-1, 2 or 3. CxF is an XML format most often used by X-Rite software. Note: CxF3 files may have extensions differnet from *.cxf. • *.xls or .xlsx: This is the standard Microsoft Excel format. It can be exported by Eye-One Share. Comma-Separated-Value (CSV) or Tab-delimited files exported from a spreadsheet application are also valid. • *.txt or other: Any other standard text based file format from any other software/spectrophotometer. Please send the file at the following address: colorchecker@babelcolor.com . In your e-mail, include any pertinent information such as the chart format, its age (edition date written on the reverse or on the product package, or purchase date, even approximate), the chart condition, and some measuring instrument info (model, illuminant, software with version number, etc.).