basic scanner questions

All talk about software, scanners, printers, digital backs, etc.
Charles Twist
Founder
Posts: 721
Joined: Mon Mar 27, 2006 6:33 pm Etc/GMT-1+01:00
Location: Cleveland
Contact:

basic scanner questions

Post by Charles Twist » Fri May 18, 2007 6:24 pm Etc/GMT-1+01:00

While out for an enjoyable visit to Saltwick bay last night, questions of fundamental importance arose regarding scanners. Dave P and I were soon out of our depth, so we resolved the online sages should have a go.
Basically, we don't understand colour management. As far as we know, there is a physicist's definition of what the three fundamental colours are, ie each of red, green and blue have an arbitrary wavelength assigned to them. The scanner's job is then to raster the surface of whatever and work out the percentage of each colour, ie how much transmission there is at each wavelength. So why the heck do we need profiling? Surely every scanner in the world knows which wavelengths to analyse. Surely every proper scanner in the world can quantify the transmission at each of these wavelengths. Why the variation?
Secondly, a good many people swear by profiling. I am quite convinced they have their sanity but I fail to see the method. Isn't it a case that there are just three colours to analyse? Why the need for such a rich palette? Is it really the case that when colours are mixed, the scanner can no longer accurately analyse the pixel? I would have thought band-pass filters to be better than that these days - I assume that is what is used to analyse the colour. I would be very impressed if each pixel were analysed for its full spectrum, although that must almost be feasible at a reasonable rate these days.
Anyway, if you have any explanations that reasonably intelligent, but scanner-ignorant folk can understand, we'd be glad to hear from you.
All the best,
Charles

Tim Myers
Forum Hero
Posts: 133
Joined: Mon Jun 26, 2006 12:43 pm Etc/GMT-1+01:00
Location: Wirral
Contact:

Post by Tim Myers » Fri May 18, 2007 8:39 pm Etc/GMT-1+01:00

Wavelength isn't the issue - it the intensity response. The sensor is not linear in its response to light, and the response it does have is subject to any number of tolerances.

My understanding anyway ;)

Tom Perkins
Posts: 34
Joined: Sun May 06, 2007 9:58 am Etc/GMT-1+01:00
Location: Cambridgeshire

Post by Tom Perkins » Fri May 18, 2007 10:40 pm Etc/GMT-1+01:00

Ok, this is a really tricky subject, and one I don't understand that well, but I think this is right:

The role of colour management is to ensure that the colours stay the same throughout your workflow, as you go from one device to another.

This is necessary because an RGB number that your camera or scanner gives you is simply a percentage of a red, a green, a blue, and you therefore need to use different percentages to get the same actual red, green or blue from different devices, like your printer, because each device interprets the percentages in a different way.

A profile is a file which tells a given printer/scanner, etc, how to accurately display colours. The output of two devices is compared, and then the profile is produced to compensate for the difference in each devices way of interpreting the RGB data, by changing the numbers.

Joanna Carter
Founder
Posts: 1283
Joined: Wed Mar 29, 2006 7:26 pm Etc/GMT-1+01:00
Workshop Images: http://grandes-images.com/fr/Paysages/P ... _2009.html
Location: Plestin-les-Grèves, France
Contact:

Post by Joanna Carter » Sat May 19, 2007 12:49 am Etc/GMT-1+01:00

Hi Charles

Not all light is the same colour, neither can the sensitivity of devices be guaranteed to be accurate, unless you pay a small fortune :cry:

Imagine you are taking a picture of the same scene, once under a bright blue sky in the middle of the day, and again, under an evening light... Well, some colours will photograph the same, some will change. The role of a profile is not just to apply an overall correction ala 81B, but to ensure that all the colours in the scene get adjusted to a standard value so that, whether you look at the picture on the monitor, scan it in or print it, you are always getting the same appearance and colour rendition.

There is nothing so expensive or wasteful of ink and paper than trying to match a print to an original tranny without a correct profile. Sure, you could get pretty close but profiles ensure that WYSIWYG.

Of course, there's more... lots more, but that is the essence.
Reassure yourself - stroke an Ebony

Charles Twist
Founder
Posts: 721
Joined: Mon Mar 27, 2006 6:33 pm Etc/GMT-1+01:00
Location: Cleveland
Contact:

Post by Charles Twist » Sat May 19, 2007 10:00 am Etc/GMT-1+01:00

Sure. I have often enough been told that the different parts of the workflow need adjusting and that is the purpose of the profiling, but what is bugging us is the variability. Are you telling me that there isn't an industry standard for red, green and blue, so that when the scanner scans red, the monitor shows red and the printer prints red? Are you telling me that no one manufacturer cannot ensure that each and every scanner they produce is exactly the same? As Dave P pointed out, when Audi knocks out one red car after the other, they're all the same red.
Joanna said:
The role of a profile is not just to apply an overall correction ala 81B, but to ensure that all the colours in the scene get adjusted to a standard value
So you are saying that unless the first two colour channels are nil (colours absent from the tranny's pixel), the scanner is unable to read the third channel correctly. I am amazed.
Thanks,
Charles

Joanna Carter
Founder
Posts: 1283
Joined: Wed Mar 29, 2006 7:26 pm Etc/GMT-1+01:00
Workshop Images: http://grandes-images.com/fr/Paysages/P ... _2009.html
Location: Plestin-les-Grèves, France
Contact:

Post by Joanna Carter » Sat May 19, 2007 10:32 am Etc/GMT-1+01:00

The problem is that things like the light source in a scanner and either the phosphors in a CRT or the flourescent light source in an LCD monitor vary in their rendition of colour temperature.

Flourescent lamps are especially poor at yielding a true white light, usually presenting a significant green spike in their spectrum. Individual flourescent lamps can vary within the same manufacturer and the spectral output of a lamp can vary over time and with age. CRT monitors are no more immune from age-related change as the phosphors can change characteristic with exposure to the exciting radiation.

Scanners tend to use either LED or cold cathode illumination and these are also subject to change.

Profiling is not meant to be a "once and for all" fix for these problems, otherwise manufacturers would pre-profile their products before shipping and your dream of perfect colour matching would be possible. Instead, we can expect to have to profile and re-profile devices on a regular basis to ensure the consistency expected of a digital workflow.

Unfortunately, too many parameters in both manufacturers' light sources and even daylight are subject to change and regular profiling allows us to maintain a consistency in the workflow from scanner, through monitor, to final print.

Real world labs that produce colour images will also use daylight matched lighting in viewing areas so as not to be affected by variations in natural daylight; they will also go to the extent of ensuring that their computer background and the walls of the room are 18% grey so that ambient reflections and objects do not influence the perception of colour. You will be amazed how your perception of what colour is correct can be affected by something next to what you are viewing.

Oh, and different papers and inks also affect the printed result, sometimes quite drastically, so that is why we need to profile the printer for each and every ink/paper combination.

All said though, you could forget profiling and produce images that are "hand-crafted" or "individual", guessing at what should be the correct colour for each print and achieving a digital equivalent of the darkroom hand print where variation is expected. But then you have to consider how much paper/ink you would waste achieving an acceptable colour balance every time you produce a print.
Reassure yourself - stroke an Ebony

Tim Myers
Forum Hero
Posts: 133
Joined: Mon Jun 26, 2006 12:43 pm Etc/GMT-1+01:00
Location: Wirral
Contact:

Post by Tim Myers » Sat May 19, 2007 1:43 pm Etc/GMT-1+01:00

Charles Twist wrote:As Dave P pointed out, when Audi knocks out one red car after the other, they're all the same red.
They're not. Your brain and eyes just make you think they are.
My father used to design spectrophotometers, scientific instruments used for the precise measurement of colour. There's a lot more to this than you think.

There is no international standard for the wavelengths of colours. Whilst it's easy to say that red light is approximately 650-720nm, can you define what you use as primary red? As most of our scanners will use a CCD either provided with light at a specific wavelength, or an element filtered to that wavelength you are at the mercy of the material yuo use to filter.

The bandpass of the material is likely to vary batch to batch (Although the wavelength of LEDs, by their very nature can be precisely controlled), the density of the material used for filtration, and the intensity of the light emitted by an LED or other source can vary wildly (And be affected by environmental factors such ast the temperature of components). The analogue stage in the scanner is equally as affected by manufacturing and environmental tolerences. Also remember that your colour film varies in colour rendition and response from batch to batch. Remember that your 30cm ruler is never actually 30cm long!

Every piece of measuring equipment needs to be calibrated if you want an accurate result from it, and your scanner is simply that - a measuring device.

DJ
Site Admin
Posts: 360
Joined: Thu Sep 07, 2006 1:48 pm Etc/GMT-1+01:00
Location: Norfolk

Post by DJ » Sat May 19, 2007 2:17 pm Etc/GMT-1+01:00

Oooook. This is a can of worms. Here goes.

First a very simple cut down explanation. When we do colour in digital, despite the fact we use hard numbers ( 255 red, 255 green, 255 blue etc ) we are not talking absolute colour (i.e. wavelengths), but percentages of colour. e.g. 255 red = 100% red. Now what is 100% red to one device, is not necessarily equivalent to 100% red on another device. If you had a car and a motorbike both going at their top speed, they could both be said to be going at "100% speed", doesn't mean they're travelling at the same speed though does it? Different devices have different capabilities. So, when one device says "100% red", the other device says "ok" and passes it on to another device, who says "ok", and what we get is the "chinese whispers" effect, where the accuracy of the original message is lost. Colour Management is the process of defining what those percentages actually are for each device, and thus providing a method for converting one to the other.

Imagine you're a fisherman, and you catch a magnificent fish, and you want to tell everyone, they ask you "how big was it?" and you say "it was 28!". Twenty-eight what? mm? cm? inches? lbs? kg?! That's the problem that Colour Management fixes, it's a universal translator for colour.

Now a bit more in-depth, forgive me if I cover stuff people already know.

What you suggest Charles about red/green/blue being known defined wavelengths of colour is partly true ( red 565nm, green 530nm, blue 420nm are the wavelengths of the "average" eye's receptors), however every other "colour" in the spectrum is also a known defined wavelength. Light itself is not made up of red/green/blue light, we don't have a red, green and blue sun, but those primaries are how we detect light, that's what's in your eye, red, green and blue sensors (cones). It's worth pointing out that there is no such thing as black or white in the spectrum, white is how we perceive all light mixed, and black is an absence of light. This is because "colour" is really a perception, it's an "event" between three participants, light source, object & observer, to quote Bruce Fraser. Colour is essentially subjective.

Scanners, cameras and film do not measure light at arbitrary wavelengths, they have grain, photosites or sensors with a coloured filter on them, and they measure light intensity at that point, not colour. The response of that filter is manufactured and calibrated to perform in a certain way for a given lightsource. The wavelength or colour of those filters will also vary from manufacturer to manufacturer, from unit to unit, there's no such thing as a perfect filter, or if there is, it's prohibitively expensive to make. The light sources too are different from model to model, unit to unit etc., they also degrade or change over time.

There are devices which can measure photons at different wavelengths, these are called Spectrophotometers, they're expensive to make, and these are what we use to create profiles. The Gretag Macbeth Eye-One spectro, which is about the most popular in photography, can measure light between 380 - 730 nanometre wavelengths, with 10nm reporting precision, that's from ultra-violet all the way to infra-red. It has a reading aperture of 4.5mm, that's not much!

Colour Management has two stages, calibration, and profiling, the two are often confused as one, and shouldn't be. Calibration is setting up the device to perform to the best of it's abilities. Profiling, is accurately measuring what those abilities are.

Two types of profiles, input ( scanner, camera etc ) and ouput ( printer, monitor etc ).

For input profiles, we measure a known colour value in a fixed light. For example, we have a target, which has been accurately measured ( with a spectrophotometer ), so we know what the value of the colour is, then we measure how the device sees it, and we then now how to convert it to it's proper value. By doing this for a range of colours about the spectrum we gain an understanding of how that device sees colour, a map of it's abilities which we can use to convert it's input. Because each scanner will have a slight different light-source ( not to mention sensor capability ) this ability will vary, and so one scanner may be able to scan or "see" more red than another.

For output profiles, we send a known value ( LAB mode ) to the device, and measure what it outputs ( with a spectrophotometer ). This way we can map the devices capabilities, so we know what number to send the device to achieve a certain colour, and what the limits of that device are. Then we can at least have consistent output, and will have some correlation with the input. It also allows us to intelligently "map" input colours to an output device, say if the input far exceeds the capability of the output device ( quite usually ), then we can map to the abilities of the device. For example, in Relative Colorimetric rendering, we can map what is 100% red to what the device's 100% red is, in Absolute Colorimetric rendering, we map that 100% red into the closest actual colour the output device can achieve, if it's beyond the gamut of the device, it's clipped.

You've been using Colour Management for longer than you think, daylight or tungsten balanced film? That's Colour Management, just done at the lab by the film designers. And the reason that Audi churns out red cars that all look the same ( even if they're not exactly ), is that they use colour management :wink: The paint mixers use spectrophotometers to ensure the colour is more or less the same in paint batches. They're even used in orange juice, to make sure it's a consistent "orange" colour, I kid you not! :D

George Hart
Forum Hero
Posts: 115
Joined: Thu Aug 24, 2006 9:32 am Etc/GMT-1+01:00
Location: Chester, UK

Post by George Hart » Sat May 19, 2007 6:56 pm Etc/GMT-1+01:00

I'm happy to stick with B&W…

Joanna Carter
Founder
Posts: 1283
Joined: Wed Mar 29, 2006 7:26 pm Etc/GMT-1+01:00
Workshop Images: http://grandes-images.com/fr/Paysages/P ... _2009.html
Location: Plestin-les-Grèves, France
Contact:

Post by Joanna Carter » Sat May 19, 2007 7:00 pm Etc/GMT-1+01:00

George Hart wrote:I'm happy to stick with B&W…
You obviously don't know what you are missing :wink: or maybe you do :roll:
Reassure yourself - stroke an Ebony

Charles Twist
Founder
Posts: 721
Joined: Mon Mar 27, 2006 6:33 pm Etc/GMT-1+01:00
Location: Cleveland
Contact:

Post by Charles Twist » Sat May 19, 2007 7:11 pm Etc/GMT-1+01:00

This is great stuff! Thank you for all your patient replies.
Joanna: to my knowledge, the light source in liquid cristal displays is a light emitting diode, and little to do with the fluorescent lights which decorate our ceilings (which are indeed green).
DJ and Tim: Yes I am aware of the subjectivity of sight and one of the wonderful things I have learnt through LF and general photography is to step back and look more coldly at the colour. I now realise that colour analysis as done by scanners is approximate in order to keep the cost down. Good on them! Calibration is indeed a cheaper alternative.
DJ, you say:
For input profiles, we measure a known colour value in a fixed light. For example, we have a target, which has been accurately measured ( with a spectrophotometer ), so we know what the value of the colour is, then we measure how the device sees it, and we then now how to convert it to it's proper value. By doing this for a range of colours about the spectrum we gain an understanding of how that device sees colour, a map of it's abilities which we can use to convert it's input. Because each scanner will have a slight different light-source ( not to mention sensor capability ) this ability will vary, and so one scanner may be able to scan or "see" more red than another.
The scanner operates in a very similar way to our eyes, detecting a narrow bandwidth of wavelengths. The relative intensity of each detector (eye's cone cells or light-sensitive diode) is then resynthesised additively. The brain has its own colour management and white balancing mechanisms of which we are more or less aware. Scanner profiles act in a similar way. That's fine.
Because each channel's detector (R, G and B) is selective for a narrow bandwidth of wavelengths, the detector perceives only the light which is emitted from the source with a corresponding wavelength. Even though the scanner's light source covers a broad spectrum, the nature of the detector is such that the light source is almost tantamount to three monochromatic light sources.
The variation in intensity from one emission wavelength to the next can be annuled by internal calibration, which would not cost any more than the existing detector (assuming the detector is white balanced). What I see as being more of a problem is the band-pass filter. If the transmission curve varies from one scanner to the next or over time, then indeed calibration is of paramount importance. And if the band-pass doesn't totally cancel out the wavelengths beyond the R-, G- and B-defining areas, then indeed you need a complex target.
How come this band-pass thing cannot be solved? Possible solutions: (i) provide genuine monochromatic light sources. It is now possible to manufacture diode lasers down to the low 400s nm. Otherwise, LEDs would not be a bad option if less light is not a problem. Then you can continue to use approximate band-pass filters without worrying about the tolerance. I don't know whether the emission wavelength varies with the age of the components, but I would have guessed it's pretty slow. (ii) Improve the actual band-pass filter. I remember about 9 years ago working with a Raman detector, where the light was detected through a filter, the transmission of which was controlled by an electrical current or voltage. Are those things still prohibitively expensive? (iii) Analysis of the optical path at the factory is not an option on account of the variation over time of the various elements' responses.
Anybody any idea whether any of these avenues are being investigated?
And finally, since you touched on it, allow me to push my curiosity further: you mention that neither white or black are detected as such. They are inferred from the amounts of R, G and B (similar to human vision although we have rod cells to help us). How does the scanner know what 100% transmission should be on each channel? That is to say, if your target shows a neutral density with 20% of each channel, how does the scanner know it's not dealing with pure white, given that it has an equal amount of each of the three channels? Does it have an internal calibration?
Thanks for your interesting comments,
Charles
PS it wouldn't be an "orange" juice if it weren't orange... :wink:

Joanna Carter
Founder
Posts: 1283
Joined: Wed Mar 29, 2006 7:26 pm Etc/GMT-1+01:00
Workshop Images: http://grandes-images.com/fr/Paysages/P ... _2009.html
Location: Plestin-les-Grèves, France
Contact:

Post by Joanna Carter » Sat May 19, 2007 8:05 pm Etc/GMT-1+01:00

Charles Twist wrote:Joanna: to my knowledge, the light source in liquid cristal displays is a light emitting diode, and little to do with the fluorescent lights...
Sorry Charles but you are under a misapprehension there. LCDs are made of thin film transistors that change their opacity with current; they have to be backlit in order to filter the light so that you can see the image. Take a lupe to that new-fangled screen you've got :wink:
Charles Twist wrote:How does the scanner know what 100% transmission should be on each channel? That is to say, if your target shows a neutral density with 20% of each channel, how does the scanner know it's not dealing with pure white, given that it has an equal amount of each of the three channels? Does it have an internal calibration?
It is certainly calibrated (sort of) but these things are manufactured in too large a quantitiy for them to be individually and accurately calibrated so, I guess, they "calibrate" to a set value which may or may not be totally accurate. And believe me, it doesn't take much to upset colour balance :roll:
Charles Twist wrote:PS it wouldn't be an "orange" juice if it weren't orange... :wink:
Blood orange anyone ? :P
Reassure yourself - stroke an Ebony

Apple
Forum Hero
Posts: 361
Joined: Sun Apr 23, 2006 7:04 pm Etc/GMT-1+01:00
Location: West Yorkshire

Post by Apple » Sun May 20, 2007 7:13 pm Etc/GMT-1+01:00

Charles Twist wrote:PS it wouldn't be an "orange" juice if it weren't orange... :wink:
...and concentrated orange juice is lime green (and toxic to wildlife) at "industrial" strength - go figure as the Americans say... :wink:

As Joanna said, equipment will be designed and assembled to a price - manufacturing tolerances are involved and you pay more for smaller ones - the colour sensor may have small tolerance but when the chain of tolerance is calculated, you may be able to "drive a bus through it" metaphorically. It's how much can things wander before you pick up a change - resolution and subjectivity (you hope to eradicate the latter by using a non-human measuring device :wink: )

DJ put it nicely and it links in with my mechanical engineering side for calibration etc.

Give something a known signal and measure the response - this is the calibration. Once you have the measurements across the range, you can apply corrections to take that output and modify it to the output it should be. i.e. the same as the input signal. This is the profile. Just because something is calibrated, it doesn't mean it's correct - you just know how far out it is - it's up to you to decide whether to do something about it to make it "right."

As for checking the calibration's right, that's where calibration to higher standards and traceability comes in. For metrology stuff, the general rule was that something had to be calibrated with a known standard that was at least 10x higher resolution and ideally traceable to national standards.

I had a quick squint at the Edmunds Optical page and their LED light sources are ±25nm - would you or the scannner notice from one to the next being potentially 50nm apart at a worst case?
Full Member of the Tearoom Appreciation Society - affiliated to UKLFPG.

Charles Twist
Founder
Posts: 721
Joined: Mon Mar 27, 2006 6:33 pm Etc/GMT-1+01:00
Location: Cleveland
Contact:

Post by Charles Twist » Sun May 20, 2007 9:03 pm Etc/GMT-1+01:00

Thanks Joanna for correcting me. Learn more every day.
Andrew, I would say that 50nm in the light's wavelength should be fairly easy to spot. 25nm in ideal conditions should be feasible. Next time I'm in a photophysics lab, I'll try it out...
Charles

User avatar
Thingy
Forum Hero
Posts: 578
Joined: Wed Aug 20, 2008 11:13 pm Etc/GMT-1+01:00
Location: Orpington, Kent

Post by Thingy » Wed Sep 24, 2008 9:48 pm Etc/GMT-1+01:00

Just to complicate matters, if when using digital you use the RAW format, as I do, the actual image is monochrome, but the software (Photoshop CS3) interprets it as colour. I shall avoid an detailed explanation. I suggest to really understand the process, just in RAW alone, you read Real World Camera RAW with Adobe Photoshop CS3 by Fraser and Schewe. They have a whole chapter on the subject.

http://www.pearsoned.co.uk/Bookshop/det ... 0000259853

Post Reply