Mathematically right values in decomposing to LAB ?

gimp

#41

I photographed the gray surface of the SpyderCheckr in the box. This is the RAW file that I obtained
IMG_0260.CR2 (20.6 MB)
Unfortunately, the L* varies a lot between the center of the white circle (supposed to be the place where the product will be) and the edges…!!
To be a usable tool, my impressions are that the set up of the box and the lighting should be redone from scratch… :confused:
Reducing the area where I will put the orange peels could be an option to get more accurate results, but then the measure of the Lab* thanks to analysis of photo looses a lot of interest to me. My goal was to be able to analyse the color of all the peels belonging to one “batch” in 2 or 3 photographs.
In addition, in this set up it means that I am not able to photograph the SpyderCheckr with constant lighting.
What do you think ?


(Alan Gibson) #42

Within the main rectangle, the lightness varies, but the chromaticity (the a* and * channels) is fairly constant.

(Contrary to my previous suggestion, I used the white balance from the camera.)

L* varies between about 10% and 19.9%.
a* varies between about 48.5% and 50.3%.
b* varies between about 49.5% and 50.4%.

(On this scale, a* and b% are neutral at 50%.)

The variation in a* and b* is local, rather than global. This suggests the chromatic variation is from the card, not the lighting.

The variation in L* is more global, with a range of 10%. In the OP, you say:

Unless the variation in the L* of the oranges is massively more than 10%, I’m afraid the results won’t be satisfactory.

If your subject was flat, like a document, you could adjust the L* values to compensate for uneven lighting. But your subjects are spherical, so this adjustment can’t be done.


#43

The variation of the L* due to differences in maturity between 2 different types of colors is likely to be about 15 or 20, but anyway the “calibration” cannot be done properly I think… :cry:


(Alan Gibson) #44

Can you share a photo or diagram of the box? A diffusion screen between the lights and oranges will help. Perhaps the lights can be rearranged.

Have you considered using a ring flash? I have no experience, but I expect this gives a constant illumination, provided the distance from camera to oranges is sufficient.

See what other people say. My experience is limited.


(Glenn Butcher) #45

I think this is an important observation. If you’re looking to measure the specific spectral response of a material with the available tools, I think you need 1) consistent lighting in both temperature and coverage, 2) take a relatively small patch from the image consisting entirely of the material in question, and 3) apply a ‘smoothing’ (gaussian blur?) function to it to average out.
Then: do your LAB conversion.


(Elle Stone) #46

I did once try a home-made ring flash (my husband is a genius at creating such things). The illumination was incredibly diffuse and even, totally too diffuse for the particular thing I wanted to photograph, no shadows at all. Anyone else have experience using a ring flash? I was making a macro shot of a small pine cone, so no idea how it would work for other setups.

Copy tables typically use two long light fixtures (perhaps high CRI fluorescents) angled at 45 degrees from the item to be photographed, one from each side. The book “Light Science and Magic: An Introduction to Photographic Lighting” has excellent instructions and diagrams for photographing just about any conceivable surface. It’s available through Amazon etc, and even my local library had a copy so it’s not hard to find - highly recommended for anyone who wants to know more about proper lighting for various purposes.

I looked at the raw file using GIMP via the darktable raw plug-in, and tried to compare “Y” (from XYZ) values to the nominal values the chart is supposed to have. I used a modified version of babl and GIMP (many thanks to Pippin and Massimo: https://gitlab.gnome.org/GNOME/babl/issues/22) and used Levels (had to adjust both exposure and black point) to set the brightest and darkest patches along the bottom to match values given in ArgyllCMS “SpyderChecker24.cie”, relying on information given here regarding the gray patches on the back of the target chart: https://www.datacolor.com/wp-content/uploads/2017/05/SpyderCheckr_24_UserGuide_EN_v1.pdf

Here’s a screenshot:

Here’s the cie values from the ArgyllCMS “ref” folder:
SpyderChecker24.cie.zip (609 Bytes)

FWIW, I agree with what @snibgo and @ggbutcher said - the current setup doesn’t really seem sufficient for the kinds of measurements you want to make. It might be worthwhile to make your own lighting setup. But if you do, I think you’d still need to put the items to be photographed in the center of the frame, maybe one or or most three or four product samples at a time, to keep lighting as consistent and even as possible and avoid light fall-off from the camera and lens, which probably accounts for a fair amount of the light fall-off from center to edges of the chart as photographed in the raw file.


#47

49

The square in the middle is in plastic (I thought it was frosted glass until I had access to it)

A photo taken from above, (the “ceiling” of the box where the camera is usually on was removed)
IMG_0280.CR2 (19.8 MB)


#48

In my posts, I have tried to provide hints to direct the conversation in certain directions, which may or may not be relevant to our conversation. E.g.,

A quick web search brings me to the following. We might be able to glean something from what people have done previously.

1. https://onlinelibrary.wiley.com/doi/pdf/10.1111/jfq.12245

image

2. https://www.researchgate.net/profile/Luiz_Pereira21/publication/322245226_Predicting_the_ripening_of_papaya_fruit_with_digital_imaging_and_random_forests/links/5a4dfb750f7e9b8284c5a28c/Predicting-the-ripening-of-papaya-fruit-with-digital-imaging-and-random-forests.pdf

image


(Elle Stone) #49

In the articles I linked to in the LCH color palette article, doing a quick check, many used spectrophotometers. However, this study of phlox colors used a scanner, which nicely solves the problem of setting up a light box, but does likely require making a custom scanner profile and of course finding a decent scanner:
https://etd.ohiolink.edu/!etd.send_file?accession=osu1429694764&disposition=inline

Thank you @afre for directing the discussion to what’s already been successfully done using digital images rather than spectrophotometers!

Doing a quick scan through of the articles you linked to, one of the studies did use a light ring. One used a black surface and black lining to the box surrounding the item to be photographed. In the setup that @MarionGaff1 is using, it almost looks like there’s a bottom bright ring around the edges of the box, which surely would contribute to flare in the camera.


(Alan Gibson) #50

If only a couple of days can be given to the problem, perhaps photograph only one orange at a time, with all oranges in the same place, and take measurements from the small area parallel to the camera sensor. Set the camera on manual exposure, the same for every orange.

If considering a new lighting setup, remember that oranges are 3D objects. If data is taken only from front of each orange, a small area parallel to the camera sensor, then the lighting should be flat, as for photographing documents. But if measurements are taken from other parts of each orange, a more ideal setup might be a light sphere with oranges in the centre and constant illumination from all directions.

A light tent might be used. This is thin white material, lit from the outside, with a hole for the camera lens.

Photos of faces taken with ring lights show heavy falloff (shadows) at the sides of the faces, so again measurements will be consistent only for small areas.

Are the lights separately dimmable? If so, that might help.

Perhaps translucent screens can selectively reduce the light.


(Alan Gibson) #51

Another thought: to test a lighting setup, spray-paint some oranges the same shade of gray. Photograph these gray oranges. Ideally, the resulting photograph will be entirely one shade of gray, with no “roundness” visible. In practice, the lighting across each orange will vary.

The more I consider the problem, the more I think the solution is one orange at a time.


(Elle Stone) #52

I think the product that’s to be photographed is actually orange peels, not whole oranges. So using a scanner isn’t as silly as it might otherwise sound. But unless @MarionGaff1 's university already has a really good scanner and there’s room in the budget for a scanner target chart, this isn’t a practical option. Personally I don’t have any idea what to look for in scanners, though surely there are such things as scanners that have suitable light sources and can accurately capture colors.

Regarding a light tent, such a thing can be improvised at a relatively low cost using a large cardboard box and tracing paper or white nylon cloth - diy instructions are readily available on the internet. But the problem with light tents is that they make such a uniform light that color in the resulting photograph tends to be washed out. Look at all the wine bottle and clothing product images (one can buy person-sized light tents for making really bad clothing catalog images) where the photographer has traded sharp specular reflections and directional lighting for a washed out sheen over the entire image.

This article very nicely shows the trade-off with photographs of tomatoes with increasingly large soft overhead lighting - imagine what the tomatoes would look like in an actual light tent:

I’m not just speaking theory (“theory” defined as someone else’s advice on how to photograph something :slight_smile: ) wrt to light tents. Quite awhile ago my husband built me a set of “tinker tubes” (https://www.photigy.com/topic/diy-equipment-using-pvc-pipe), and the first thing I did was set up a light tent (edited: originally I typed “light box” instead of “light tent” - sorry!) and take some photographs, and with some amount of dismay realized the trade-off between rich colors and surrounding an object with uniform diffuse lighting.

The “tinker tubes” are marvellously useful for many things including setting up backdrops and directing/modifying light sources. But for the truest richest colors a “dark tent” surrounding the item to be photographed works much better than a “light tent” - in one of the articles @afre linked to notice the set-up did use “dark” surround and that seems like a good approach to photographing orange peels.

But with a “dark tent” - which can be a large cardboard box painted flat black inside or covered in black cotton flannel or etc - and with the light sources inside the dark tent, one is now back dealing with the problem of reflective glare and evenly lighting the product. In many cases (specular reflections to one side), the solution is appropriate light sources above and to either side placed at a 45 degree angle to the direction the camera lens is pointing, such as the arrangement for copy stand lighting.

Changing topics, mention was made of using a blue background that can easily be separated from the orange color of the orange peels. I’m wondering if a black background might be better - light bounces around an awful lot and I’d be leery of somehow introducing a blue color cast - in a dark tent this wouldn’t be such a problem, but at the other extreme of a light tent, it seems possibly something to be concerned about. Maybe compare the exact same shot (lighting setup and such) of an orange-colored paper sample with a black vs a blue background, and of course see if the black background can be extracted. But surely it can, but extracting backgrounds isn’t something I have any practical experience with.


(Alan Gibson) #53

Ah, yes, orange peel not whole oranges, so ignore most of what I wrote. Thanks Elle. A good scanner could be a solution, if they exist.


#54

I found a scanner Epson Perfection 2580 Photo, it is a bit old but it is made for scanning negatives. Plus, it comes with a Adobe ICC profile.
https://epson.com/Support/Scanners/Perfection-Series/Epson-Perfection-2580-Photo/s/SPT_B11B172091#drivers
Do you think it could be suitable?

Do you mean that using the SpyderCheckr24 will not work as a “scanner target chart” ?


#55

I own a scanner that is not-too-bad. I have a Nadorcott in the fridge. Would it help this discussion forwards if I were to peel it, scan the peels, and publish the scan?

Sincerely,
Claes in Lund, Sweden


#56

Thanks Claes!
Actually, if you own a piece of uniform gray paper or PVC that you could scan, we could see if the resulting scan has uniform L* a* b*. (or even better a target chart ?)
At the moment, I did not manage to find an adaptator for the Epson scanner…
If the results are not good with your scanner then it is not worth spending hours trying to find the good adaptator or another scanner…!

But you can keep your Nadorcott and eat it later on :slight_smile:
Thanks a lot


#57

When it is all said and done, could someone send me a vial of essential oils. :rofl:


#58

I just tried, @afre, but gmail refused to handle it :frowning:


(Elle Stone) #59

Hmm, this is a case where I spoke without having any practical experience, something I don’t like to do because too often I end up saying stupid stuff :slight_smile: .

Upon reflection, the answer is "I don’t know, but I think maybe the SpyderCheckr24 would be just fine. Here’s why:

Wolf Faust makes affordable and well-regarded target charts for both cameras and scanners: http://www.targets.coloraid.de/

The “C1” chart is for cameras. The scanner charts are specific to the type of film or paper print that will be scanned. This thread talks about the huge difference getting the right film/paper-specific target can make in results of scanning: http://www.largeformatphotography.info/forum/showthread.php?38435-New-IT8-target-for-my-V750&s=bcfad935c9f1eab2bb34e56618007591

But you wouldn’t be scanning film or a print on paper. Instead you’d be using the scanner as a camera. So to answer your question, if I had to guess, I’d guess that maybe your camera SpyderChecker24 target might work just fine because you’d basically be using the scanner as a camera. But this is just a guess. If you have the time I’d be really curious as to how your resulting ICC profile might work when applied to things you scan, starting with the target chart.

Anyone have any experience using a camera target chart to profile a scanner that’s being used as a camera? @gwgill - is using a camera target in this situation a reasonable thing to do?

Regarding the Epson scanner you mentioned, I just don’t know. All I have access to are old online reviews. I tried to find mention of that particular scanner being used to scan artwork - a case where color accuracy is very important - but didn’t find anything. I did find a shop that sometimes uses a scanner for digitizing artwork, but they didn’t say which scanner, and they did say they prefer using their very pricey large format camera: https://www.bellevuefineart.com/fine-art-scanning-process/

If I had to give advice, and please note again that this advice isn’t based on my own practical experience, I’d say between your camera and your scanner the camera is the better choice. But it doesn’t hurt to try making a profile for both camera and scanner, and compare results.

If you decide the camera does seem like the better option, then I’d suggest setting up a dark tent in the form of a large cardboard box painted black or lined with black paper, or draped with black cloth - black cotton flannel would work just fine. I’d use the target chart that you already have, and I’d try to find out whether the lights that you already have are the CRI>90 bulbs. If not, I’d look for better lighting.

Is there an Art Department at your university (or maybe a neighboring university’s art department?) that might have good lights that they’d let you use, and perhaps even a flatbed scanner suitable for scanning fine art?

Edit: even if some other university department has a flatbed scanner, and even if they use that scanner for scanning artwork, doesn’t mean their scanner would be better for your purposes than your camera and a dark tent combined with good lighting. The scanner by definition has its own light source(s), and so quality of the generated scanner light will affect the accuracy of resulting colors.


#60


Have fun!
Claes in Lund, Sweden