How to test if a print was made at 16 bit?

[Aram Hăvărneanu]

Good grief, this thread is posted faster than I can type.

Doug is saying that by having 'Use Dither' checked, the 16 bit pipeline is invalidated.

Why is it invalidated? Dithering (and noise shaping) is an essential step of digital filters, especially for bit-depth changes. It is unclear why Adobe even lets you turn it off. Perhaps for scientific applications.

99.9% of photographers have 'Use Dither' checked, as this is the Adobe default. Therefore, when they say that it is impossible to distinguish a 16 bit print from an 8 bit print, they are making that judgement based on incorrect Photoshop settings.

From a signal processing point of view, having dithering on is the only "correct" setting.

Is it possible or not to tell if a print was made (I don't know the proper word(s) to use here, so Ill' try) from a 16 bit image file?

If dithering is correctly applied (ON both in Photoshop and internally in the printer driver and printer hardware), no, you can't tell.

If dithering is not applied (OFF both in Photoshop and internally in the printer driver and printer hardware), yes, you can tell very easily.

If dithering is not applied in Photoshop (OFF) but it is applied internally in the printer driver and printer hardware (ON), maybe you can see it, maybe not.

Not so, according to Doug Gray, who earlier in the thread said: 'Just turning on the dither option eliminates any advantage 16 bit driver paths provide which is mostly theoretical in any case.'

Noise shaped dithering allows to encode signal amplitudes below the naive noise floor of N stops of dynamic range, where N is the bit depth. So yes, dithering allows to encode signals of less than 1 bit of amplitude with an 8 bit signal, so less than 8 bits of amplitude with a 16 bit signal. It's a similar effect to increasing bit depth. Dithering does this at the expense of noise, but the assumption is that printer noise dwarfs dithering noise.

There may indeed be dithering happening in the print path alone and we can't control that.

Yes, almost certainly it has to be. The printer head/mechanics are driven by an analog signal that is generated by DACs on the printers. The digital input to these DACs is not the image file, it's some other kind of digital signal derived from the image file (either in the driver, or in the hardware itself). This signal processing involved in generating these digital signals most certainly applies dithering (and very likely involves bit depth conversions) because there is no reason why it would not.

[digitaldog]

Good grief, this thread is posted faster than I can type.

Why is it invalidated? Dithering (and noise shaping) is an essential step of every digital filter, especially for bit-depth changes. It is unclear why Adobe even lets you turn it off. Perhaps for scientific applications.

Yes! Those of us that use Photoshop for colorimetric testing need such an option. And more to the point, not all software products that can print photos provide this option or control. So unless we know the OP is only using Photoshop and only cares about this so called 16-bit print path, then we either need to address or ignore it's use.

From a signal processing point of view, having dithering on is the only "correct" setting.

Except when it isn't. 

If dithering is correctly applied (ON both in Photoshop and internally in the printer driver and printer hardware), no, you can't tell.

Dither can be applied, on some kinds of data in Photoshop. I don't believe Photoshop has any such control over the print driver (which driver?). And even if they do, we have no idea if the dither is the same when used to make say colors space conversions which is WHY we see the option where it resides.

The key word above and below it s IF!

[Doug Gray]

Good grief, this thread is posted faster than I can type.

LOL!

Why is it invalidated? Dithering (and noise shaping) is an essential step of digital filters, especially for bit-depth changes. It is unclear why Adobe even lets you turn it off. Perhaps for scientific applications.

Right. it's not invalidated, it's just superfluous. The printer driver does its own dithering when it comes up with a recipe for a small set of inks.

From a signal processing point of view, having dithering on is the only "correct" setting.

Right. The only disadvantage is that the image can be less compressible because it has higher entropy. However, as a practical matter, there is usually little difference since photo images normally have significant bit noise in them already. Mostly the size difference shows up when converting synthetic images.

If dithering is correctly applied (ON both in Photoshop and internally in the printer driver and printer hardware), no, you can't tell.

If dithering is not applied (OFF both in Photoshop and internally in the printer driver and printer hardware), yes, you can tell very easily.

If dithering is not applied in Photoshop (OFF) but it is applied internally in the printer driver and printer hardware (ON), maybe you can see it, maybe not.

Noise shaped dithering allows to encode signal amplitudes below the naive noise floor of N stops of dynamic range, where N is the bit depth. So yes, dithering allows to encode signals of less than 1 bit of amplitude with an 8 bit signal, so less than 8 bits of amplitude with a 16 bit signal. It's a similar effect to increasing bit depth. Dithering does this at the expense of noise, but the assumption is that printer noise dwarfs dithering noise.

Well stated.

Yes, almost certainly it has to be. The printer head/mechanics are driven by an analog signal that is generated by DACs on the printers. The digital input to these DACs is not the image file, it's some other kind of digital signal derived from the image file (either in the driver, or in the hardware itself). This signal processing involved in generating these digital signals most certainly applies dithering (and very likely involves bit depth conversions) because there is no reason why it would not.

[digitaldog]

The printer driver does its own dithering when it comes up with a recipe for a small set of inks. 

And we have no control over that dither. Is it different if someone sends 16-bit or not? How to know?
The dither option in PS is a red herring rabbit hole. Unless again the OP wants some answer that only applies to Photoshop. And someone can demonstrate what's really happening which is the crux of this discussion as the hole gets deeper.
Why not make two prints, look at em and move on?

[felixpq]

I quoted you just so you know I'm trying to talk to you. Someone said you replied with something about dark gradient. Can you or someone else give me the number of the post, just to make sure I have the right one, thanks.

[elliot_n]

Someone said you replied with something about dark gradient.

Post #21

[faberryman]

Are you getting more acceptable answers to the same question over on DP-review?

They did show one image where banding was visible on an 8-bit print but not a 16-bit print, so why not print in 16-bit and avoid the issue, even if it rarely comes up? There doesn't appear to be any downside. How much effort does it take to check a box and then not worry about it?

[Doug Gray]

This has to be one of the most pointless, in a practical sense, discussions ever. No one here has ever claimed, or can even point to anyone, that sees a difference between printing a 16 bit, non-synthetic, photo image using either an 8 bit path or 16 bit path at the driver.

None the less it appears to be an interesting technical discussion.

So, in honor of Bill Atkinson, I have created an image I call Bills Black Ball. The reason that Bill's set of color gradient balls were created was because Bill felt that banding was easier to see in circles rather than linear gradients.

The attached tiff image, in ProPhoto RGB, has RGB values ranging from RGB(20,20,20) to RGB(35,35,35) at the outer edges. It is quite suitable for demonstrating 8 bit or higher paths for both monitors and printers.

For Monitors:
View this in Photoshop with the background set to black and full press the 'F' key so the screen menus don't show. Banding on 8 bit displays should be quite obvious. It's even more obvious if you use a display setting at max cd/m^2 and/or if you make a special monitor profile using gamma=1.5 or lower. It's a good way to tell if your Monitor/CPU/Display driver is actually producing 30 bit image results.

For Printers:
Print with a 8 bit driver using Null-Transform or RelCol on a glossy type paper or anything with Dmax of 2 or higher. Make sure the Edit->Photoshop->Dither option is off. It's best to print this w/o using color management using the Null-Transform technique but it can be printed normally too. The banding will be more regular with the Null Transform but will occur with both. You should then trim the white borders of the print and examine the print in a bright light. The banding will be quite obvious.

This can be repeated with the Dither selected and all or most all of the visible banding will disappear.

[digitaldog]

They did show one image where banding was visible on an 8-bit print but not a 16-bit print, so why not print in 16-bit and avoid the issue, even if it rarely comes up? There doesn't appear to be any downside. How much effort does it take to check a box and then not worry about it?

+1 and none.

[Doug Gray]

They did show one image where banding was visible on an 8-bit print but not a 16-bit print, so why not print in 16-bit and avoid the issue, even if it rarely comes up? There doesn't appear to be any downside. How much effort does it take to check a box and then not worry about it?

+1 and none.

Yup. I just wouldn't pick a printer based on whether it supported 16 bit or 8 bit printer paths. It's got to be one of the most insignificant specs ever.

[faberryman]

Yup. I just wouldn't pick a printer based on whether it supported 16 bit or 8 bit printer paths. It's got to be one of the most insignificant specs ever.

It looks like the Canon, Epson, and HP photo printers support 16-bit, so you don't really have to make a choice, other than to check the box.  And it looks like with the Windows versions you don't even have the choice not to.

[Aram Hăvărneanu]

Now that we have what to print next thread should be about how to test if a scan of that print was really made at 48 bit or not! Then we need to print the scan again, just to make sure.

[Mark D Segal]

Now that we have what to print next thread should be about how to test if a scan of that print was really made at 48 bit or not! Thrn we need to print the scan again, to make sure.

Just to anticipate that discussion, what if I told you my 48-bit scanners actually uses 64-bit when the infra-red channel is activated?   

[elliot_n]

This has to be one of the most pointless, in a practical sense, discussions ever. No one here has ever claimed, or can even point to anyone, that sees a difference between printing a 16 bit, non-synthetic, photo image using either an 8 bit path or 16 bit path at the driver.

None the less it appears to be an interesting technical discussion.

So, in honor of Bill Atkinson, I have created an image I call Bills Black Ball. The reason that Bill's set of color gradient balls were created was because Bill felt that banding was easier to see in circles rather than linear gradients.

The attached tiff image, in ProPhoto RGB, has RGB values ranging from RGB(20,20,20) to RGB(35,35,35) at the outer edges. It is quite suitable for demonstrating 8 bit or higher paths for both monitors and printers.

For Monitors:
View this in Photoshop with the background set to black and full press the 'F' key so the screen menus don't show. Banding on 8 bit displays should be quite obvious. It's even more obvious if you use a display setting at max cd/m^2 and/or if you make a special monitor profile using gamma=1.5 or lower. It's a good way to tell if your Monitor/CPU/Display driver is actually producing 30 bit image results.

For Printers:
Print with a 8 bit driver using Null-Transform or RelCol on a glossy type paper or anything with Dmax of 2 or higher. Make sure the Edit->Photoshop->Dither option is off. It's best to print this w/o using color management using the Null-Transform technique but it can be printed normally too. The banding will be more regular with the Null Transform but will occur with both. You should then trim the white borders of the print and examine the print in a bright light. The banding will be quite obvious.

This can be repeated with the Dither selected and all or most all of the visible banding will disappear.

So this test demonstrates that it's a bad idea to turn off 'Use Dither' in Adobe's Colour Settings when printing a 16 bit file to an 8 bit print pipeline? Hmm, I'm not sure if anyone would have even considered turning it off unless you had brought it up.

With 'Use Dither' turned on, prints from an 8 bit and 16 bit pipeline look the same - no banding.

And prints using the 16 bit pipeline loos identical, regardless of whether 'Use Dither' is on or off.

[Mark D Segal]

..................
And prints using the 16 bit pipeline loos identical, regardless of whether 'Use Dither' is on or off.

........which confirms my above-mentioned interpretation of what that option applies to - 8-bit only.

[Doug Gray]

Just to anticipate that discussion, what if I told you my 48-bit scanners actually uses 64-bit when the infra-red channel is activated?   

I've actually looked at the effective bit depth of 48 v 24 bit reflective scans on the V850. Turns out there is only a slight gain with 48 bits however if one does multiple scans and averages the result some additional bit depth can be discerned. On average individual pixels in an 8 bit channel have rounding noise that is about twice that of the same pixels in 16 bit channels where photon noise dominates for an effective bit depth of 9. Averaging 4 scans together and comparing them to another 4 scans reduces the pixel variations by about 40% which is just under 10 effective bits.

However, there is considerable variation from one pixel to another that seems to be due to slight variations in the CFAs and this is on the order of an lsb of an 8 bit channel. So only a little is gained since the CFA effect is stationary and not improved by averaging. However, I don't know if this applies to transmission scans.

In any case I see no value in 48 bit reflective scans v 24 bit ones. Higher DPI scans are another story though there is significant rolloff above 300DPI on reflective scans though it's quite gradual. It requires sharpening for scanning images with fine detail to recover the detail.

[digitaldog]

In any case I see no value in 48 bit reflective scans v 24 bit ones.

Until you begin to edit the scans after scanning them (if you do). Again, the added precision is about editing the data, if the data is edited.

[Doug Gray]

........which confirms my above-mentioned interpretation of what that option applies to - 8-bit only.

Right. The dither option affects conversions from different 8 bit colorspaces as well as conversion from 16 bits to 8 bits in the same colorspace. It doesn't affect 16 bit to 16 bit conversions at all.

It appears to quite a smart "dither."  For instance if you have an RGB value of (127.25, 127.25,127.25) in 16 bits and convert to 8 bits it will populate with approx. 25% (127,127,127) and 75% (128,128,128) rather than just adding a low level of noise. It's likely it does the exact same thing when converting to an 8 bit printer driver.

[Doug Gray]

Until you begin to edit the scans after scanning them (if you do). Again, the added precision is about editing the data, if the data is edited.

Agreed about the editing which should always be done in 16 bits if possible. One can introduce significant artifacts editing in 8 bits. Especially if one has a heavy hand. It would be interesting technically to try editing both an 8 bit scan and a 16 bit scan but initially converting the 8 bit to 16. Not sure any visible difference would show up. But it's also a pointless exercise since just scanning to 16 bits/ch is simpler.

[elliot_n]

They did show one image where banding was visible on an 8-bit print but not a 16-bit print, so why not print in 16-bit and avoid the issue, even if it rarely comes up?

I'm not so sure.

This is their example (Doug's example?):

https://www.dpreview.com/forums/post/61793580

It is not clear how it was printed.

I can only induce banding like that if I print a 16 bit gradient with either:

1. Adobe Color Print Utility, with 16 bit unchecked

or:

2. Photoshop, with 16 bit unchecked, AND 'Use Dither' unchecked (in Photoshop's Color Settings)

Neither of these printing methods represent a set-up that would be used to make a normal 8-bit print. In a real-world scenario Photoshop would be used, '16 bit' would be unchecked, and 'Use Dither' would be checked. The resulting 8 bit print would be indistinguishable from a 16 bit print.