Photographic Reference Scale (2019) | Hacker News Reader

Photographic Reference Scale (2019)

smallpond.ca

99 pointsmattowen_uk35 comments

This seems very handy – however the reference image would be infinitely easier to print on the correct size if it was a PDF file. PNG image format does not retain the resolution IIRC.

For anybody wondering – the correct DPI is about 508 for some reason.

508 pixels/inch = 200 pixels/cm, exactly. The resolution unit in PNG is either the meter, or it is unspecified. 1 inch = 25.4 mm, 25.4 * 2 = 50.8. 508 is close enough to 600, and 600 dpi is a resolution commonly found in desktop printers.

If you've worked with PNG files under Photoshop, you may have noticed that a picture saved with exactly 72 dpi has 72.009 dpi when you close it and open it again. Using an integer multiple of 254 for the dpi prevents such a drift.

It's a nice round metric measurement. 508 dots per inch = 20 dots per mm.

With a PDF file you can also make the image crop box be exactly the reference image area, add crop marks and put the CC note and other info possibly inside or outside the crop area.

Its all very well saying "here's a free photographic reference, download it and print it". But I worry about the lousy job your average cheap home inkjet printer will do of getting it onto paper.

Some things are just better bought from a reputable manufacturer than bodging your own. Photographic reference charts may well be one of them.

Printing middle grey with an uncalibrated printer is uh... kinda meaningless.

https://smile.amazon.com/WhiBal-White-Balance-Pocket-Card/dp...

I've been using this for my photo adventures; I try to keep it in my wallet at all times (I worry about it getting discolored by rubbing against other cards though) and whenever I notice that the white balance is gonna get fucky (esp when there are multiple light sources) I whip that out and take a picture of it in the same scene.

edit: Also people love to use grey cards for getting exposure right but it's complete bunk science; if you're using the raw workflow (if you're fine tuning white balance you almost definitely are) you should shoot as bright as practically possible and not too bright to prevent clipping and adjust exposure as you see fit in editing.

Thanks for posting the link, because I was about to ask. It seems like printing out the reference on a home printer would make the accuracy dependent on whether or not your printer was well calibrated (and then you're back to the same problem about color accuracy.)

Wouldn't practically all printers do middle grey with only black ink/toner/whatever, thus having very little chance of having any color cast? Considering how color temp adjustment is not very exact thing to start with, I feel that generic printout should be good enough for most uses. And for color critical work, you'll want a full color swatch (eg colorchecker) anyways instead of just middle grey.

I don't know the impact of this, but:

- there is no such thing as perfectly black ink

- color printers will mix color ink into greyscale printouts

A "fun" workaround would be to print a very high resolution black-and-white checkerboard. Printers generally have plenty of resolution to make that appear as gray even to the naked eye. Of course managing to print something like that without something in the pipeline trying to anti-alias stuff and introduce gray back is left as an exercise to the reader.

Unless you are doing extreme precision professional work--likely an extremely small subset of the people reading this, if any--a calibration target from a black and white laser printer or a photo printer is fine and will give great results.

Some photographers seem to worship color correction in a weird way. Other error factors, like reflected light hitting your calibration target, are likely to supercede any minor color imprecision from those kinds of printers.

I was scanning satellite photos yesterday and came across what I think are ground reference points built to enable determination of exact ground distances using a series of concentric rings and oriented linear features. I believe knowledge of the exact diameter of the rings and length of the chords would allow a user to measure features in the photo with acceptable precision.

I know that early pilots had a series of markers that they could use to keep themselves on track as they flew their long distance routes. These things were oriented concrete markers, some of which are still accessible.

We all need a reference, a way to cross-check or ground truth our results and conclusions.

After reading this article I noticed the reference on this BBC article: https://www.bbc.com/news/science-environment-57853537 (first picture)

Though I note that that one looks suspiciously like the contra-recommended laser printed output.

Baader-Meinhof phenomenon - or did reading that Beeb article inspire someone to find and post this one?

Thanks, this is handy.

I’ve done some little experiments but having a standard-ish thing is better.

Would this integrate to give scale, in say, Meshroom? Or it’s manual calibration step someplace along the workflow?

I’m in it more for the reasonably accurate 2d/3d measurements, but approximately accurate color is nice as well.

I'm curious if there are any good methods for correcting lens distortion. Like can I print a poster, take a photo and use that to create a correction map?

If I remember right, the professional software "DXO Optics Pro" was doing this. Since then, they bundled that functionality into their newer products:

https://www.dxo.com/technology/lens-distortions/

The way it works, iirc, is you can choose the model of your lens and the software fixes things. Works for professional lenses, unsure about others.

Opencv has basic distortion correction: https://docs.opencv.org/4.5.3/dc/dbb/tutorial_py_calibration...

For more advanced stuff, look up CAHVORE models and related models.

Does anyone know of an opencv which can be used to calibrate calibrate using something like this?

Is a single gray really enough to calibrate color?

Not all color, but you don't need all color. You need color temperature and maybe luminosity, and it's good for that.

> You need color temperature

For artificial light (or mixed sources, e.g. from reflections), it is not correct. The light source is characterized by its whole distribution (which can be anything) rather than one temperature parameter (for the Planck distribution).

An easy example is a fluorescent lamp spectrum (https://commons.wikimedia.org/wiki/File:Fluorescent_lighting...).

You're going to have a hard time calibrating that without a whole bunch of patches with known reflectance/transmission profiles, in practice you just have to color-correct scenes under such light by eye and there will be things that look wrong. When I did this professionally I'd still use the grey as a primary reference along with skin, since we're very attuned to what those should look like.

> You're going to have a hard time calibrating that without a whole bunch of patches with known reflectance/transmission profiles

If you care about accurate color, you do use a card with "whole bunch of patches" of different colors. For example, these British Library guidelines on photographing and scanning archival material say:

"An industry-approved colour chart or greyscale reference card is essential in order to check the accuracy of the colour captured in the photographs. This is a physical card that is available to purchase on the internet. It cannot be downloaded from the internet and printed, or inserted into your image using software. Approved colour checkers that have been used in EAP projects include: Gretag Macbeth ColorChecker, QP cards, Kodak Color Control Q13 card, Past Horizons Bookmark Photo Scale."

(https://eap.bl.uk/sites/default/files/Guidelines%20for%20pho...)

ColorChecker (“Macbeth”) Chart: https://poynton.ca/notes/color/GretagMacbeth-ColorChecker.ht...

QP cards: https://www.argraph.com/QP202.html

Kodak Q13: https://www.kodak.com/en/motion/page/color-separation-guides...

Bookmark photo scale: https://www.pasthorizonstools.com/Photo_scale_bookmark_p/boo...

Thanks, forgot what they were called. Maybe the state of the art has moved in the ~10 years since I got out, or it's just the part of it I was in (fine art, prints for the magazine to reproduce (also forgot what those were called...)), but correcting to a color checker was manual and usually inexact anyway so I'd cue off of the grey and whatever thing in the image had a tone we'd have a strong expectation of how it should look (again, usually skin).

As you quote they get used in archiving but those also tend to be done under well-characterized lighting and if not, it's something that's worth spending the time to tweak every little thing.

It's a long shot from being able to apply automatic perspective and grey-point corrections plus a little bit of tweaking so the colors look "right" (or as right as possible under the wonky spectral distributions of most lights these days).

I did manage to convince the company to buy the transparency so we could calibrate the film scanner's output.

There’s a wide variety of photographic color meter light calibration charts, see https://www.amazon.com/gp/bestsellers/electronics/3109910011...

My favorites are from Datacolor or X-Rite.

Calibration indeed presumes shooting a multi-color target. I personally use the process[0] from DCamProf docs.

[0] https://torger.se/anders/dcamprof.html#workflow_dcp

> skin

Is there a standard skin color or something? How do you deal with different skin colors?

No standard color of course, but we're so attuned to looking at people that we internalize certain ideas of what skin looks like (or should look like), more so than other objects out in the world. The big things I'd avoid are cooler tones (blue, green) as well as yellow; a touch of red or magenta is tolerated much better. In fact because there's no standard skin tone, that provides latitude to get something that looks right without nailing it exactly as long as the pitfalls I mentioned above are avoided.

Unfortunately, cameras only capture RGB values, not a whole spectrum. It is impossible to perfectly color-correct the image unless you know both the spectrum of the light source and of the materials in the scene. A reference card with lots of colors still only provides you with the known spectra of the colors on that card, it still might not contain a color with the exact same spectrum as that flower you are trying to photograph for example.

Another issue is that even if you have the perfect photography + post-processing setup, when I see the result it might be on a non-calibrated monitor, or on paper but in a room with non-perfect lighting, and perhaps I was just looking at something else a second before and my brain's auto white-balance hasn't caught up yet. So I might still not perceive it as was intended. This is why that gold dress looks blue to some people.

The patch of grey on the reference scale + good lighting of the scene to begin with will get you 99% of the way, the remaining 1% is extremely hard to get right, if at all possible.

Well, in principle you get 3 degrees of freedom instead of 1 (though, I would love to see camers with more than 3 color filters).

In practice you do 2 anyway (with exposure correction being used even more often than white balance). So I agree that more than that is usually not worth the effort.

From my amateour photography experince - with strange lighting you cannot do much anyway.

With multiple light sources, the white balance is undefined and you need to compromise somehow, yeah.

Even a scene where there is strong sunlight and a bright blue sky will make white balancing quite difficult, as the shadows are lit with sky blue skylight and the sunlit parts are lit with the sun.

I use darktable and it's possible to correct the blue tint in shadows, but like... it's annoying. And at some point I'd say fuck it and just accept that photography is not a science.

No, but in this case, the best is enemy of the good.

That, and the card also has white and black.

Neutral gray is normally used for exposure.

The theory behind modern light metering is that if you average the "brightness" of every pixel (or the average density of crystals in the case of film) is should "visually" be around 50% brightness or density, or neutral grey. Spot metering on something that's already neutral gray sort of short cuts that process

Reply on news.ycombinator.com