HomeUpSearchMail
NEW

Using the below ideas for optimization it is possible to do color optimization of pictures that incorporated a color/gray scale in general.
In most instances thought the IFRAO standard scale is being used as an example. The purpose of the IFRAO standard scale is multi-functional:

• know the size of the object in study (so IFRAO standard scale is 10 by 4 cm),
• provide a means to focus the photographic equipment (by means of black/white blocks)
  
• provide a cheap color card (cheaper then the existing commercial ones, like; Profile Prism or IT8 [~$15], Macbeth ColorChecker [~$70], Munsell Soil Color Chart [~$120 - 190] or Kodak [~$21]); the IFRAO standard scale is free for AURA/IFRAO members
  
• provide a standard to be able to do color calibration/optimization (to compensate for color errors from the moment the picture is taken until and including its publishing).

The first three aims are of course well met by the IFRAO standard scale.
Because I personally find the last item the most important/interesting for the IFRAO standard scale, I wanted to study that by looking at the following points, so that optimization/calibration can become a scalable possibility (say if one wants to do 100's of pictures) for researchers:

• determine the sRGB/AdobeRGB/Lab color values for the two versions (1994 and 2001) of the IFRAO standard scale, because these were not available [pers. comm. R. Bednarik [2003]).
  
• find/write a (shareware) program that can optimize the pictures (that include an gray/color scale) in an objective and reproducable manner and which can be gotten in the public domain (does not have to be free;-).
  
• evaluate the effectiveness of the IFRAO standard scale and if needed; propose improvements for the next version of the IFRAO standard scale.

These ideas were presented at RASI 2004 in Agra India on November 30th, 2004 (11:30-12:00).
Furthermore pictures with color scale have been tested free for people of the ROCK-ART at ASU.EDU community.

These three items will be looked at in the following sections (remember most of the below text is independent of the IFRAO standard scale, but for any gray/color scale in a picture):

• sRGB/AdobeRGB/Lab reference colors
• IFRAO standard scale
• Macbeth ColorChecker
  
• My methodology
• Software
• Optimization using Paint Shop Pro 8
• Optimization using PhotoshopCS
  
• Definitions of functions
• Evaluation
  
• Things to watch for when making/optimizing a picture
• Software improvements
• Comparing different methods
• Free optimization service
  
• Space for improvement of IFRAO standard scale
• Small changes
• Somewhat bigger changes
• Future steps
• Acknowledgments

sRGB/AdobeRGB/Lab reference colors

IFRAO standard scale

Don't use a printed version of this picture as the scale on the rock art!
Only use the official IFRAO standard scale, because that one is printed under controlled conditions.

• Bold values are out of gamut.
  
• *Reflection Density (RD) values come from: Digital color re-constitution in rock art photography, Bednarik, R.G. and Seshadri, K., Rock Art Research 12: page 42-51, 1995.
  The values at Gamma=2.2 for the gray patches are determined by using the method seen on this web site: code_grey=255*10^(-RD/2.2)
  and then this page is used for converting from say Adobe RGB (Gamma=2.2) towards sRGB values.
  The RGB values (Photoshop 4 did not have any defined RGB workspace) come from: La calibración computarizada a color en las fotografías de arte rupestre, Bednarik, R.G., 2002, http://mc2.vicnet.net.au/home/auraesp/shared_files/spanish.pdf
• **sRGB/AdobeRGB values are based on CIE Lab/D50 values (2° standard observer, measured by Joseph Wilhelm using Spectrocam for one IFRAO standard scale of May 1994). This page is used for converting towards xRGB values (using D50)
• ***sRGB/AdobeRGB values are based on CIE Lab/D65 values (2° standard observer, measured by Joseph Wilhelm using Spectrocam for one IFRAO standard scale of Nov. 2001, two other readings for different scales were done with values within 1 DeltaE [so not really visible by eye]). This page is used for converting towards xRGB values (using D65)

Macbeth ColorChecker

My methodology

• Bring the non linear input dependent RGB file (e.g. by scanner, digital camera, etc.) into a linear RGB space by determining a device dependent Gamma function.
• In this linear space, transform the values of each RGB channel by means of 3x3 RGBmatrix.
• Then do a fixed Gamma of 2.2 for the whole picture. This coincide with the ideas in C. Poynton The Rehabilitation of gamma (page 8, except that the Gamma used on this web-page is 1/Gamma in his document).
• These values are best fitted with the specific xRGB reference colors.

Software

• Excel spreadsheet with Solver add-in and the ColorOpt excel file. To get the latest version, please provide why you want to use it and acknowledge that all feedback will be shared with the author, who can publish it on the web.
  If the Solver is not yet in your Excel, then install it: Tools -> Add-Ins... -> click Solver Add-In.
  Excel Solver performs most of the times oke, alternatively another add-in can be used: Evolver.
  
• A paint/picture program, like:
  
• Jasc Paint Shop Pro (this program is the easiest to use)
• Adobe PhotoshopCS
  

Optimization using Jasc Paint Shop Pro 8 or 9

• Gamma_psp = Gamma
• Channel mixer_psp = RGBmatrix
  

1. get the picture (below one is a picture made with digital camera) with gray/color scale into your drawing program:
   
   
2. Determine for each of the color patches the average RGB values (using three times the PSP Dropper Tool (which in itself averages upto 11*11 pixels) or Photoshop Eyedropper Tool (which in itself averages upto 5*5 pixels), and calculate the overall average using the Input worksheet). One can also use the averaging function of areas in the programs (Photoshop: Filter -> Blur -> Average or PSP: Adjust -> Blur -> Average...).
   Take these areas of the scale that have no glare.
   
3. Input these average values into the coloropt20xrgbextern.xls excel spreadsheet (RGBmatrix Opt worksheet: area C6:J8 or when calculations done in Input worksheet; copy Sample database worksheet A2:I4 to RGBmatrix Opt C6:K8 with Paste Special... -> Values)
4. It is possible to toggle each color patch (area C10:J10); to define if a color patch will (=1) or will not (=0) take part in the optimization.
5. Start the solver by: Tools -> Solver... -> Solve ([Ctrl-r] and OK])
6. The above determines the optimum Gamma_psp and Channel mixer_psp.
7. Have Paint Shop Pro in 24-bit color space (Image -> Increase Color Depth -> 16 Million Colors (24 bit) Ctrl+Shift+0])
   
8. Use the value in C38:C40 for all channels in Gamma_psp (Adjust -> Brightness and Contrast -> Gamma Correction... [Shift+G])
9. Fill in the RGB values from G38:I40 in the Channel mixer_psp (Adjust -> Color Balance -> Channel mixer...)
10. Use the value 2.2 for all channels the Gamma_ps (Adjust -> Brightness and Contrast -> Gamma Correction... [Shift+G])
11. A script is available (use right click: Save [Link] Target As...)
12. This gives the optimized picture:
    
    
    Which one can compare with this picture:
    
13. The whole above process can be done in some 10 minutes per picture.

Optimization using Adobe PhotoshopCS

• Curve_ps = Gamma
• Channel mixer_ps = RGBmatrix
  

1. Do the same as the above steps 1 to 5
   
2. The above determines the optimum Curve_ps and Channel mixer_ps.
   
3. Have Photoshop in sRGB mode (Edit -> Color Settings [Shift+Ctrl+K]-> sRGB IEC)
4. Activate 16 bits/channel in Photoshop (Image -> Mode -> 16 Bits/Channel)
   
5. Fill in the values from D38:E46 in the Curve_ps (Image -> Adjustments -> Curves... [Ctrl+M])
6. Fill in the RGB values from G38:I40 in the Channel mixer_ps (Image -> Adjustments -> Channel mixer...)
7. Fill in the values from J38:K40 in the Curve_ps (Image -> Adjustments -> Curves... [Ctrl+M]) for doing a Gamma=2.2
8. No scripting is yet made by the author for Photoshop.

Definitions of functions used in optimization method

Gamma

Evaluation

Things to watch for when making/optimizing a picture

• For general ideas about important things for rock/megalithic art photography see this link.
  
• No glare on the gray/color scale. Even light objects (like a flash, sun but also a clouded sky) can sometimes be seen in the scale. So be sure that the surface of the standard scale is in such a direction that (if it were a mirror) one would only see dark objects.
• If there is glare in the gray/color scale, try to use the parts of the color patches that have no glare in it.
• Use the bare color/gray standard scale (so no protective covers, etc.), otherwise colors will be captured wrongly.
• Have the scale uniformly lit by (in)direct light.
• Have the scale standing as straight (water level) as possible in the picture (so that one can select the maximum rectangular color area for optimization).
  
• Obey the rule of the scale having closer then the maximum distance. E.g. The IFRAO standard scale should not be used further away then 1.5 m with normal focal distance (55 mm at 35 mm [photo film camera] or 8.5 mm at 1/2.7 inch [digital camera]).
  For other focal distances; make the long side of the scale never smaller than 1/8 of long side of the picture. This makes the selectable even-colored color/gray areas (except black) around 35*15 pixels.
  
• Use files that are as pure as possible, at least loss-less; like tiff, bmp, etc.
  
• Even is one does its evaluation (error determination) in Lab space one still has to check if the RGB values are within the below mentioned boundaries. The RGB values are the actual values in tiff and bmp files, so these are the important values to check (not the Lab values).
• Try to have as less as possible RGB channel values close to 0 or 255. If close to these values (could be due to under or over exposure), evaluate and see if one has to remove such a color patch from the optimization.
• Try to have the maximum amount of contrast. If one can not analyse ones picture on site, make a few with 0, -0.5 ,  -1 or 1.5 time-stop (over)exposure for relatively dark colored objects or 0, +0.5, +1, +1.5 time-stop (under)exposure for relatively light colored objects.
  
• If the standard deviation of RGB values over the averaged samples for each color patch is zero (could be due to under or over exposure), evaluate and see if one has to remove such a color patch from the optimization.
• If the average standard deviation of RGB values over the averaged samples is high (perhaps scratches, different lighting conditions, wrong sampling, compressing errors (like in jpeg), etc.), evaluate and see if one has to remove such a color patch from the optimization.
• If the scale is in places damaged, don't use these places to determine the color patch values.
• Check that a darker patch is indeed darker than its lighter brother. If not, evaluate and see if one has to remove such a color patch from the optimization.
• If one does not want to have the color scale on the picture, make two pictures under the same lighting conditions (so fast after each other). One picture has the color scale in it (so just press the shutter of the camera), then one presses the shutter halfway down (on most cameras this will fix the configuration of it, like whitebalance, speed, etc) and while keeping it halfway pressed, remove the color scale and make thesecond picture. Now one can optimize the color scale values from the first picture resulting in a profile. that can be used for the second picture.
  

Software improvements

• ICE Color
  
• The proposed procedure tries to do the optimization process of ICE Color in (semi-)linear space. The problem thought is that the Gamma is unknown of the initital picture (a value of 2.2 is assumed, which is normal for most xRGB color spaces); that is why the step 4 of the procedure uses a Gamma of 0.45 (=1/2.2). In some cases this could make the behavior of ICE Color inconsistent.
  The other disadvantage of the above procedure is that at each rendering (F5) the color space gets 8 bit/channel deep and thus loosing valuable dark color information (color noise).
  
• inCamera
• Needs to get the IFRAO standard card colors in the program. For this a business case needs to be set up. It is important to check if 8/16 color patches would produce a proper optimization (ICE Color is not able to handle this low amount of colors properly).
  
• PSP
• A script is available for doing the whole methodology.
  
• Increase color depth per channel to 16 bits (or higher)
  
• Have the cursor depicting the size of the Dropper Tool
  
• Allow user to define him/herself the size of the Dropper Tools
• Photoshop
• Have the cursor depicting the size of the Eyedropper Tool
  
• Allow user to define him/herself the size of the Eyedropper Tools
• Have a published gamma function in PhotoShop (and not only in ImageReady)

Comparing different methods

• In this analysis the IFRAO standard scale is handled as a given. If one wants to make (small/big) changes, see the appropriate section.
  
• It does not make sense to do this work if something already exists, so I have asked IFRAO if they have a method for doing the calibration, and they say that no tool is in the public domain for this (pers comm. B. Bednarik [2003]).
• All pictures had the IFRAO standard scale (1994 version) and the Macbeth ColorChecker card in it (the original pictures made by Robert Mark).
• All the pictures were handled in the sRGB color space (because computer screens and thus web page viewing are in this color space).
• The Lab values were gotten by using Photoshop (Window -> Color -> Palette Options... -> Second Color Readout -> Lab Color). Be sure to have Photoshop in sRGB color space (Edit -> Color Settings [Shift+Ctrl+K]-> sRGB IEC).
• When the method-id has the letters MB at the end; the optimization is against the sRGB Macbeth ColorChecker colors, if no letters are appended the optimization is against the sRGB IFRAO colors.
• Remember that an error (DeltaE) of around 1 is just perceptible with the naked eye.
• The average error (DeltaE) when comparing random colors with the reference colors is around 75.
  

• VR
  Using the above devised transform method with the IFRAO color patches for optimization
  Software needed: Microsoft Excel, coloropt20xrgb.xls, Paint Shop Pro (~$ 89) or Photoshop (~$ 599)
• VR/MB
  Using the above devised transform method with the Macbeth ColorChecker patches for optimization.
  Software needed: Microsoft Excel, coloropt20xrgb.xls, Paint Shop Pro (~$ 89) or Photoshop (~$ 599)
  
• ICEColor
  Using ICE Color with the IFRAO color patches for optimization
  Software needed: ICE Color (~$ 30)
  
• ICEColor/MB
  Using ICE Color with the Macbeth ColorChecker patches for optimization
  Software needed: ICE Color (~$ 30)
  
• inCamera/MB
  Using inCamera with the Macbeth ColorChecker patches for optimization
  Software needed: inCamera (~$ 150) plus PSP (~$ 89) or Photoshop (~$ 599)
• BGW
  Using black, gray and white (eye)droppers in Image -> Adjustments -> Levels (Photoshop, incl. Options...-> 3xTarget Colors) or Adjust -> Color Balance -> Black and white points... (PSP, incl. 3xDesired Color) with black, gray and white patches of the IFRAO card
  Software needed: Paint Shop Pro (~$ 89) or Photoshop (~$ 599)
• BGW/MB
  Using black, gray and white (eye)droppers in Image -> Adjustments -> Levels (Photoshop, incl. Options...-> 3xTarget Colors) or Black and white points... (PSP, incl. 3xDesired Color) with black, neutral5 and white patches of the Macbeth Color Checker card
  Software needed: Paint Shop Pro (~$ 89) or Photoshop (~$ 599)
  
• EnhancePSP
  Using the Enhance Photo option in PSP 8. This method does not really look at a color scale, so no optimization in the strict sense can be expected.
  Software needed: Paint Shop Pro (~$ 89)
  
• iCorrect
  Using the iCorrect method. This method used a Memory Color for each of the IFRAO color patches, in effect thought it does not really look at the exact IFRAO color values.
  Software needed: inCorrect (~$ 150) plus PSP (~$ 89) or Photoshop (~$ 599)

1. Case 1
   Where the average DeltaE (average color error) is calculated referring to the IFRAO color scale in the picture for four different light conditions: sun, shade, fluorescence and copystand light
   
2. Case 2
   Where the average DeltaE (average color error)  is calculated referring to the Macbeth color scale in the picture for copystand and fluorescence light.
3. Case 3
   Here we take only one row or some specific colors (like equivalent with IFRAO colors and an softer alternative colors) from the Macbeth ColorChecker scale to optimize (so not all). The average DeltaE is calculated referring to the remaining Macbeth color patches in the picture.
4. Case 4
   Here we change the number of color patches that take part in the optimization.
   

Case 1

Case 2

In case 2 we use the same methods, but now the Lab/D65 average DeltaE (CIE 1994) is referring to the Macbeth color patches.
The results of the methods that were not optimized against the Macbeth Colorchecker card (to minimize dependancy/correlation) are:

My own method (VR) does not have the best numerical results, but using the naked eye it does not look that bad;-)

Orginals on left and optimizations on right
(fluorescence, cloudy, shade, sun, copystand light)

Case 3

Here the Lab/D65 average DeltaE (CIE 1994) is referred to the Macbeth color patches, while we use ICE Color for optimizing:

• each row (6 colors) of the Macbeth color scale to optimize (1st row: top row of colors, 4th row: bottom [neutral] colors) (xxx row/MB)
  
• equivalent colors as the IFRAO card on the Macbeth card (8 colors: blue, green, red, yellow, white, neutral 5, neutral 3.5, black) (Equiv. color/MB)
  
• alternative softer colors than the IFRAO card on the Macbeth card (8 colors: light skin, blue sky, foliage, orange yellow, neutral 8, neutral 5, neutral 3.5, black) (Alt. color/MB)

This gives the following result when only taking those MacBeth ColorChecker patches that are not used for determining the optimum (to minimize dependancy/correlation):

Only done for copystand light, because that light seemed to give in most
methods comparable results as the average results in case 1

Case 4

Here the Lab/D65 average DeltaE (CIE 1994) is referred to the Macbeth plus IFRAO color patches which did not take part in the optimization using ICE Color. The number of Macbeth ColorChecker patches in the optimization is varied between 0 and 24. The following results were gotten:

Comparing the cases

• It looks that for case 1 and 2 the different lighting conditions behave comparable (at least for copystand and fluorescence).
  
• The ICE Color, VR and inCamera methodologies perform comparable in case 1 and 2.
  ICE Color though is not always consistent (I am discussing this with the authors) and inCamera is not working easily for the IFRAO standard scale (see next bullet). At this moment I prefer the VR method.
  
• It would be interesting to see the behavior of inCamera when optimizing it against the IFRAO standard scale, but then a business case needs to be made towards PictoColor Cooperation.
• The method using the black/gray/white dropper tool (BGW) is not really consist and has a large variance in error.
• Interesting to see that the Black, Neutral 5, White optimization in case 4 (using ICE Color) looks better than in case 1 or 2 (using the BGW method), this could be due to different methologies.
• From case 4, it looks that the more color patches one uses, the smaller the average DeltaE and the smaller the variation in error is.
  
• The limited amount of color patches in the IFRAO standard scale will limit the color optimization/calibration that can be reached.
  
• There is not much difference in results between real colors (case 3, 1or2or3rd row/MB) and grays (case 3, 4th row/MB), so I assume that grays are just as important as normal colors for the optimization.
  
• Comparing average errors (DeltaE) in case 3 of the Equivalent IFRAO colors and the Alternative softer colors, it looks like that the hard colors provide a comparable error as the softer colors (the difference between the two is hardly significant).

Free optimization service

If you have a rock art picture with a color scale in it (Macbeth or IFRAO) you can send it to me (remove underscore in presented e-mail address), I will try to optimize/reconstruct the colors for you. But you have to do something in return;-), I will resent three low quality pictures first:

1. one using the program ICE Color (using semi linear space)
2. one using the program ICE Color (in xRGB space)
3. one using my own methodology together with PSP.

After you tell me which picture has the colors that are the closest to the original, I will send you that specific picture with approximately the same file size as the original. After doing that I will delete all the related picture files from my systems (so I will not use your files except for the above purpose).

Just to be sure, I can't give any guarantees about the quality and also my initiative is not linked to any organization; so it is just a private initiative! Due to that, I will limit my services to my own discretion.

So if you want to utilize this opportunity; send your file(s) to me (remove underscore in presented e-mail address) (please don't send files bigger than 2 MBytes, if not communicated with me first) and tell me if you want to have it optimized towards AdobeRGB or sRGB.

Space for improvement of IFRAO standard scale

• Remove the glare that can be gotten from the present standard scale (perhaps by using glare free paper?).
  
  
  

• In some cases the black ink of the black side of the card smudged the color patches and thus changing the color properties. This should be rectified.
  
  
The above smudging (at A) of the rectangular black patches is on the back of a card  gotten in RAR, Vol. 21, No. 1, May 2004 (directly photographed after receiving). At B the smudge has been erased using a pencil eraser (one can't do this on the colored side, because then the color changes). The smudging (of the backside  text) can get onto the color patches of the scale laying on top of this, changing the color properties.

• One of the Nov. 2001 IFRAO color patches (green) is marginally outside the sRGB gamut, this could easily be rectified (change the sRGB from [0,150,82] to say [20,150,82])
  
• The macro scale will incur a larger average error (DeltaE) in the optimization, due to a smaller amount of colors (6). Perhaps put the macro scale on the right side instead of the left side (or move the frame with the IFRAO text to the other side), so that at least more other patches (possibly 2 more) can come in the picture.
• I don't believe that changing the color values on the scale will change the effectiveness of the optimization (see the results between hard and soft colors).
• The accompanying paper 'Introducing the IFRAO standard scale' to the standard scale, needs to include more tips about its practical use as a color optimization/calibation tool (and the possible pitfalls and how to overcome them). Possibly creating an online FAQ (Frequently Asked Questions).
  For instance: If one does not want to have the color scale on the picture, make two pictures under the same lighting conditions (so fast after each other). One picture has the color scale in it (so just press the shutter of the camera), then one presses the shutter halfway down (on most cameras this will fix the configuration of it, like whitebalance, speed, etc) and while keeping it halfway pressed, remove the color scale and make the second picture. Now one can optimize the color scale values from the first picture resulting in a profile. that can be used for the second picture.
  I use this two picture method for reconstruction of photos made of my wife's paintings.
  
• One has to be aware that the color optimization/calibration that can be reached with the IFRAO standard scale, is limited (due to the small number of color patches), but it is in a good direction.
  
It seems that the present color patch sizes are already perceived as too small. This would increase the size of the card, I think.
But keep in mind the following with regard to the design principle of the IFRAO standard scale:
"One scale should be used for distances of up to 1.5 m. Between 1.5 and 4.5 m, two scales are required. The scale cannot be used with precision at distances exceeding 4.5 m, using lenses of standard focal length. Best results will be achieved at distances of under 1 m."
By the way, multiple scales will not really work when using software like ICE Color.

• the average DeltaE (error between theoretical color values and color values after optimization)  is still quite large with the IFRAO standard scale, due to the limited amount (8) of color patches (grays are in my opinion also colors). If one wants to improve the color scale for color optimization/calibration the number of color patches needs to be increased:
  
• making the scale larger than the present IFRAO standard scale (some people do not really prefer that)
  
• make all patches the same size (and keeping in mind that the present patches are by some perceived as too small already)
  
• change the black/white blocked edge into contrast rich colors (not really preferred, because this would minimize the focussing properties of the scale).
• Because I see the color optimization/calibration the most important, I propose the following changes towards 16 colors:
  
• change the blocked edge into contrast rich colors so that focussing is still possible (by optimizing the difference between the patches looking at Hue, Intensity and Saturation),
• keep the 1 cm gauges, this is done by combine the 1^st and 2^nd row of color, where each color is 2 cm, but shifted 1 cm.
  
• make all the patches as big as the gray patches, because there is no difference in importance between 'grays' and 'colors'
• It could be wise/handy to keep grays in the new color scale.
• Four black squares (as far as possible apart) to be able to see if glaring/uneven lighting has happened. This could also serve to compensate for uneven lighting.
  
• change the colors in such a way that they are evenly distributed over the rock-art color gamut.
  In the 1994 (and thus also the 2001) version the distribution is of colors is somewhat baised, see below:
  
  
  
• If we distribute the colors in the linear RGB color space and then use a gamma of 2.2, we get the following possible color choice of:
  
  
  The above gives the following draft color scale in sRGB (most of the white text will be removed of course;-):
  
  
• what more colors will do on the production costs of the scale, I don't know, feedback is welcome on this.
  
• perhaps increase the size of the scale (say with 50% or 100%).
• are the above measures really that big a change? Some measures could be utilized, I would say, without even changing the real character of the IFRAO standard scale.

Future steps

• See if the measured Lab/sRGB values of the IFRAO standard scale can be reproduced.
• Determine the consistency of the results from the optimization software.
• Determine what software/programming language is good to use (looking at software availability, web services, economy, target group of users, accuracy/color noise, etc.)
  
• Optimize the methods: study the effect of integer calculations (it could be that the optimization procedure and the fact that integers are used, not enough accuracy is available and thus reducing the number of unique colors).
  Perhaps higher bits per channel are needed. The visual results of PSP/python are compable with Photoshop, although PSP/python looks to have higher noise looking in the histogram (PSP [8-bit/channel], python Image package [8-bit/channel], Photoshop [16-bit/channel) and ICE Color [64-bit/channel if done in one Render, otherwise 8-bit/channel]).
• Be able to automatically determine average color values of the recorded gray/color scale (perhaps by pointing out the four corners of the card and let the program decide where the colors are).
  
• Make a web-service for this optimization methodology (perhaps using two files: one of the gray/color scale as cut from the art picture and the art picture (with the same standard scale).

Acknowledgments

Disclaimer and Copyright
HomeUpSearchMail