Converting 8 bit images to 16 bit - any benefits?

[Tony W]

In another thread I posted the following about my current belief and understanding relating to converting 8 bit files to 16 bit i.e. I doubted that any real benefits would be had and further that there may be potential to make matters worse by taking this course. Quote:
When I started using Photoshop I used to convert 8 bits to 16 bits in the belief that I would be gaining something. My revised view is that it may be possible but there are risks associated that need to be understood - I have seen/introduced posterisation into my own images which could be attributable to editing original 8 bit as 16 bit.

So now this is my current thinking. Of course I accept my thinking may be out of date, plain wrong or just a load of c**p and would welcome different views

If an image is available as anything over 8 bit then I will edit in 16 bit. This includes scanning and acquiring from DSLR. In my case I thought my DSLR was 12 bit but it is actually 12 bit compressed which I believe equates to a real bit depth of only 9.5 bits!

There is a problem when converting 8 to 16 bit which may not be seen or we are even aware of. The original 8 bit image has 256 levels when converted to 16 bit it will have 32769 (think PS is actually 15 bit hence losing 32000+ levels). This must mean that there are huge gaps in the histogram. AFAIK PS histogram of a 16 bit image has been converted to an 8 bit view therefore the gaps are probably not apparent. Depending on editing steps taken it is conceivable that the gaps will increase even more due to spreading the 8 bit level information over the new 16 bit levels?

There was some civil and polite discussion with differing views put forward; however this discussion caused a drift off topic and therefore I am starting this thread in the hope of gaining a better understanding.
Due to my comment about having experienced posterisation in the past it was requested that I post an example or two for opinion. I know that this is going to sound like a ‘cop out’ but I just cannot find the examples – I am still looking and will post if and when I find them.
As it is somewhat embarrassing to make this kind of statement and not able to show some evidence I decided that I would select an image and have a play to see if I could repeat issues I experienced by mildly abusing the image file. While I was not able to get any posterisation (probably poor choice of image due to a bland sky!) my results showed no appreciable differences between the 8 bit edit and the 16 bit edit. I do accept that the results of just one test and the conditions I chose to apply prove nothing and there are probably better ways to test and make informed conclusions. But for what it is worth this is what I did and my results.

• Original image taken with 3Mp Fuji camera - I know, bland and uninteresting but it was the best I could find in the time frame
• Duplicated the image and converted one to 16 bit
• Applied the following to both 8 and 16 bit images
• Applied levels layer and reduced Output to 100 and 155
• New levels layer and adjusted shadows and highlights to just touch each end of histogram
• Curves layer and applied an S curve

I scrutinised both images at 100% trying to find any differences - I was not expecting to see improvement over the original (cannot make a silk purse out of a sows ear!) of course but simply what differences. In particular I was looking for losses in the 8 bit that would no show up so much in 16 bit. I found nothing.

Then converted the 16 bit back to 8 bit and looked again - found no difference! The histograms in the second image tell the story. Left hand image 8 bit original histogram after the levels and curves adjustment. Right hand image 16 bit. Original histogram to the left. Right hand histogram after converting the image back to 8 bit.

I was a little surprised that I could not pick up anything between the 8 or 16 bit versions particularly as the 8 bit histogram looks so poor.

So as I have not had an epiphany with this short test I am left thinking either my test was woefully inadequate, my logic totally flawed, wrong type of image to highlight differences any combination of these.

Just so all is clear I am merely trying to establish what benefits if any from converting an 8 bit image to 16 bit for editing. This does not relate to the debates about 16 bit editing as 8 bit - in that case I am all for keeping in 16 bit for as long as possible and just cannot subscibe to the view that it is a waste of time

Anyone?

As I finish this I am just about to embark on the adventure of changing ISP's (tomorrow) - think I would rather visit the dentist. Hope I am not let down like my last change - 1 month without service!!!
Anyway I'll be back

[andrewrodney]

Converting the original data from 8-bit to 16-bit provides nothing useful for that original data. But of course, if you paste higher bit data into that newly converted data, or build gradients, edit that data severely etc, then yes, it can be somewhat useful. The original data is suck in its state. You didn’t improve it by converting it to high bit.

The bottom line is, if you start with high bit data, keep it that way.

[freezehead]

Yeah, I think there's no differences when you convert 8 to 16 bit 'cause what it does is just "interpolate" more values between 2 consecutive value of 256 values that 8 bit created

[ftp-Jeff]

No, it will do nothing to help you... Depending on the job, but I mostly process out of Capture 6 into 16 bit..... Do the whole retouch until client is happy. When it is time to supply Certified proof, I then convert to 8 bit....

[kav]

Originally Posted by andrewrodney

Converting the original data from 8-bit to 16-bit provides nothing useful for that original data. But of course, if you paste higher bit data into that newly converted data, or build gradients, edit that data severely etc, then yes, it can be somewhat useful. The original data is suck in its state. You didn’t improve it by converting it to high bit.

The bottom line is, if you start with high bit data, keep it that way.

In response to this I'm just wondering, how do you deal with 16 bpc mode when dealing with huge composites especially in cmyk with the use of smart objects? I know not everyone deals with this but some of us do and it can slow things significantly.

[andrewrodney]

I’d do as much work as humanly possible in RGB, 16-bit, then convert to CMYK, 8-bit and move on. CMYK is an output color space for a specific print condition, so one of the very last steps would be the conversion to CMYK.

[RobertAsh]

Originally Posted by kav

In response to this I'm just wondering, how do you deal with 16 bpc mode when dealing with huge composites especially in cmyk with the use of smart objects? I know not everyone deals with this but some of us do and it can slow things significantly.

I'd recommend doing some careful tests like Tony W did. If you can go through your entire workflow in 8-bit and 16-bit and see no difference in the end product to your customer then you might want to weigh whether or not for your workflow there is a practical difference.

The risk there, of course is that although you might usually be able to do the job over if you found something in rare cases, if you were in a time crunch with no time to do the job again, that would hurt. There are some workflows were 8-bit all the way through can cause banding (John Wheeler demostrated a couple of examples). So whether that applies to you workflow or not, that requires testing. I've read photographers' assessments that they don't need 16-bit. But that's their workflow, not necessarily yours.

Another alternative is to upgrade your computer - faster hard drive(s) e.g. RAID 0 or RAID 5 eSATA, more cores, faster data bus and more memory. That's what I ended up doing recently. It's definitely been worth it.

[ftp-Jeff]

Definately should work in 16 bit.... Yes file sizes can be large... But the hardware out there is more than capable of handling it nowdays.

[kav]

Originally Posted by RobertAsh

Another alternative is to upgrade your computer - faster hard drive(s) e.g. RAID 0 or RAID 5 eSATA, more cores, faster data base and more memory. That's what I ended up doing recently.

I wanted to some time ago. The 2009 mac pro update was absolutely dismal. They basically went from an 8 core with a ton of room for ram and cut that machine in half but adopted a new architecture. 2010 update was a spec bump + 6 core added at the top. Once again I don't like their choices for the dual socket model and the 6 core feels like it could outgrow its ram unless I used expensive 8 GB sticks. I'm waiting to see what they come out with this year. i'd like something that could hold everything in ram.

Currently a RAID system wouldn't do much for me unless I used a small internal RAID0 for scratch disk space. I plan to alleviate my scratch bottleneck (which isn't too terrible) with massive amounts of ram when I upgrade my computer. Disk speeds aren't so much the bottleneck most of the time in saving the massive files. I heard adobe made a change to make this easier but waiting for it to compress everything as it saves is the real bottleneck as it's a single threaded process.

So yeah I'm hoping for some speed increases at 16 bpc levels when I do see an upgrade. Before anyone mentions it I don't really like imacs.

[RobertAsh]

Originally Posted by kav

...Currently a RAID system wouldn't do much for me unless I used a small internal RAID0 for scratch disk space. I plan to alleviate my scratch bottleneck (which isn't too terrible) with massive amounts of ram when I upgrade my computer. Disk speeds aren't so much the bottleneck most of the time in saving the massive files.

Not necessarily true. Disk speeds can be the single biggest bottlenecks when reading or saving files. Or when Photoshop uses cache on disk (which is does a lot of). For large files RAID-0 can cut in half the times you spend reading and writing files.

For me, many of my files go into 100s of MB. A few are 1 GB in size or larger. So RAID 0 helps a lot.

What you may be seeing is a Photoshop quirk in its design. I found that when I doubled my memory my files save times vastly improved. Ordinarily that shouldn't be the case, but Photoshop writes a lot to and from disk. So if you don't enough memory it will clog your system with all kinds of changed data, writing to and reading from the disk cache.

But even with that out of the way (memory, yes, was the worst bottleneck in this case) my new, shorter save and read times would double if I wasn't using RAID 0.

[Tony W]

Andrew, thanks for your input. You confirmed my thinking that the data is fixed in its original state and I do appreciate pasting higher bit data or gradients should be undertaken as 16 bit. You are after all adding new data and that should be treated correctly to achieve the best quality.

The bit about editing the data severely intrigues me. Are you saying that if you need to undertake severe editing 8 bit data should be converted to 16 bit? If that is the case how severe would the editing need to be or under what circumstances do you feel 16 bit be better/safer option?

Thanks to all for the comments

[ftp-Jeff]

No 8 bit to 16 bit conversion is pointless! Start with a processed 16bit file.... Keep it that way until you have finished.

[andrewrodney]

Originally Posted by Tony W

The bit about editing the data severely intrigues me. Are you saying that if you need to undertake severe editing 8 bit data should be converted to 16 bit? If that is the case how severe would the editing need to be or under what circumstances do you feel 16 bit be better/safer option?

There is no free lunch here, and converting the 8-bit to 16-bit isn’t anything like having the original data. That said, try this. Take a 16-bit file, something like a gradient. Duplicate it so you now have two. Take the duplicate and convert to 8-bit per color. Now convert back to 16-bit. You can try running some edits on it, USM, curves etc. Run the same edits on the original 16-bit file and if you wish, on a third dupe that is only 8-bits per color.

What I think you’ll find is all three are different, but deciding on what or how useful the differences are is difficult (you want to have holes in the histogram as your criteria?). What I’ve found is that of course, the best, cleanest data is the original high bit data file. But the image converted from 8-bit to 16-bit and then edited appears a tad less “damaged“ but its difficult to say if using the histogram is a reliable gauge (are the gaps less due to the math or is there just more padding going on with that conversion to 16-bit?)

Bottom line is, don’t worry about it, stick with original high bit data and edit that. If you have 8-bit per color data, that’s what you are ‘stuck’ with.

[Tony W]

ftp-Jeff, thanks I did feel from the outset that from 8 -16 bit conversion likely to be pointless. I would always prefer to work at the higher bit rate and am generally unconcerned about larger file sizes etc. The point is that sometimes you have no control and are stuck with 8 bit.

Andrew, thanks I intend to run your little test. As to the histogram I have to say that I could not care less how it looks as long as the final image meets my criteria. I am not in the business of trying to produce attractive histograms - but was a little surprised by how poor my 8 bit test histogram looked compared to the 16 bit and the fact that I could not find any differences between the 2 images.

[andrewrodney]

Originally Posted by Tony W

Andrew, thanks I intend to run your little test. As to the histogram I have to say that I could not care less how it looks as long as the final image meets my criteria.

Criteria how and where? On a low rez display? On a halftone 133lpi press? A high quality fine art Inkjet printer?

I agree, the histogram doesn’t tell you everything about the qualities of a document. Short of printing it out to lots of different devices and looking real closely, it does tell you something about the edits.

We also don’t know what the data will produce after further editing or conversions to some kind of output color space.

[Tony W]

Point well made and taken. Generally I would be aiming to print on a good quality inkjet such as my own HP B9180 - a little old maybe (and possibly quite short of qualifying as a Fine Art Inkjet) but still IMO capable of producing excellent prints B&W and Colour

[John Wheeler]

Thanks for posting this thread Tony

All of my post is of course just IMHO

I too agree with the other posts that just converting from 8 bit to 16 bit does not add quality to the image. I don't know of any case where the image quality can be improved by just the conversion from 8 to 16 bit depth. I believe you were looking for something else in your post however:

I am merely trying to establish what benefits if any from converting an 8 bit image to 16 bit for editing.

My experience has been that converting an 8 bit image to 16 bit before doing additional processing in Photoshop can reduce the introduction of further degradation in practical situations relative to doing additional Photoshop processing while remaining in 8 bit depth. This degradation when it occurs is not necessarily noticeable in many images for two reasons:

1) The eye is quite forgiving in many cases for small changes in images. An analogy is with high quality JPG compression. There are many subtle changes in JPG compression in the image yet subtle enough that many do not notice them at all.

2) These subtle changes can also be masked by noise. The degradation effects may not be noticeable if the image contains a reasonable amount pixel noise which is common. In addition, Photoshop actual adds dither (1 bit noise) with default settings in a variety of operations including when 16 bit images are converted to 8 bit. This helps reduce the visibility of banding due to 8 bit quantization. I personally prefer to leave this option turned off and add noise/dither when I think it is needed rather than letting Photoshop put it in automatically.


There are a variety of image types using what I believe are simple non-extreme edits where I have observed image degradation when leaving 8 bit images in 8 bit mode for editing. I will show one of those cases below and post more if I have time. The anomalies do show up when printing as well.

Degradation Case 1

While the eye is forgiving for small degradation in many images, it turns out that the eye can pick up subtle luminosity changes in a field of slowly changing luminosity/color which is usually called banding. So here is a case where you will see banding degradation when an 8 bit image is processed in 8 bit mode yet not when it is processed in 16 bit mode. This occurs when this 8 bit sRGB image is converted to a working space of ProPhoto RGB.

To actually see the degradation there are several conditions that are required:
- You need an image that is slowly changing in luminosity/color
- This image needs to be a low noise image or the banding will be masked by the preexisting noise.
- You need to turn off the Photoshop dithering/noise which is on by default which would otherwise hide the banding in 8 bit mode. Turning off dithering is done by going into Edit > Color Settings and unchecking the dither option (found in "More Options"). The dithering is pretty good at hiding the banding yet you have to then live with the extra noise.
- It is also necessary to view the image at 100% magnification or greater as to not introduce display anomalies (it could show banding when there really is none)

Here is the link to the full resolution 8bit sRGB JPG sunset image I used in this example. It is a low noise version leveraged from a Stock Exchange image (details at end of post):http://s774.photobucket.com/albums/y...RGBFullRes.jpg

Here is a reduced resolution version of that image.

Sunset low noise 8 bit sRGB low res.jpg

The following image was created through these steps:
-Opened image directly into Photoshop. I was flagged that the sRGB space did not match my Working Space of ProPhoto RGB. I took the option to convert it to ProPhoto RGB.
- Leaving the image in 8 bit mode I duplicated the image twice to two additional layers and set the Blend mode to Screen to brighten it up
- Flattened image
- Edit > Convert To Profile…sRGB

This next shot is a 100% view on the left side of the image where the clouds are. You can see a good amount of banding.

Sunset-8-bit-sRGB-roundtrip-to-ProPhoto-RGB-and-Screen-Layers.jpg


Here are the steps taken to create the following image.
- Brought the same 8 bit sRGB image into Photoshop and again was flagged that it did not match my working space of ProPhoto RGB
- This time I took the option to leave it in the Working Space of sRGB
- Converted to 16 bit mode
- did Edit > Convert To Profile…..ProPhoto RGB
- Again duplicated the image to two more Layers with Blend mode set to Screen
- Edit > Convert To Profile…sRGB
-Converted back to 8 bit mode

Here is the resulting 100% magnification image. As you can see by doing the processing in 16 bit mode the banding has been greatly reduced.

Sunset 16 bit sRGB roundtrip to ProPhoto RGB and Screen Layers.jpg

Without going into the details taking the 8bit sRGB image into a much wider Gamut Color Space slightly posterized the data. The screen blends exaggerat that posterization.

This is only one of many examples of degradation when an 8 bit image is left in 8 bit mode for Photoshop processing as opposed to first converting to 16 bit.

If this type of situation is rare enough for you or the banding is already masked by noise, this may not be an issue for your images. So I don't even have to think about it, I just convert to 16 bits if all I have is an 8 bit image to work with.

If I have time, I will post more examples.


Steps to Create the Low Noise version from the Stock Exchange image
--------------------------------------------------------------------
The low noise version of the Stock Exchange sunset image was created by:
- Copy Stock Exchange image to computer. Link to Stock Exchange image: http://www.sxc.hu/photo/1347672
- Open JPG into ACR
- Luminosity noise reduction to 100%
- Luminosity detail set to 50%
- Go into Photoshop as 16 bit sRGB (this was done because ACR actually introduces dithering noise if you bring it into Photoshop directly in 8 bit mode)
- Convert to 8 bit mode (remember to have dithering option turned off in Color Settings)
- Re-save as 8bit sRGB JPG at quality level 12

[kav]

Originally Posted by John Wheeler

Thanks for posting this thread Tony

This occurs when this 8 bit sRGB image is converted to a working space of ProPhoto RGB.

Was this just for testing? I can't imagine why you'd want to convert an image from sRGB to ProPhoto. There's literally nothing to gain there. I thought it was only made an opyion because it's the space ACR uses to make corrections. The most realistic use I can think of would be in product photography. If the color of a product was out of gamut/losing detail to single channel clipping, it could make sense as a stop gap assuming it prevents single channel clipping or compression. If you're importing into sRGB though it has already clipped/compressed whatever colors did not fit within this gamut.

[John Wheeler]

Originally Posted by kav

Was this just for testing? I can't imagine why you'd want to convert an image from sRGB to ProPhoto. There's literally nothing to gain there. I thought it was only made an opyion because it's the space ACR uses to make corrections. The most realistic use I can think of would be in product photography. If the color of a product was out of gamut/losing detail to single channel clipping, it could make sense as a stop gap assuming it prevents single channel clipping or compression. If you're importing into sRGB though it has already clipped/compressed whatever colors did not fit within this gamut.

Thanks for the comment Kav. I agree that there is no gain in taking an sRGB image to ProPhoto RGB. My personal approach is to leave the image in the color space that is embedded with the image. That said, there are many books out there that tout using ProPhoto RGB for everything and there are many folks that have Photoshop set up to with their default Working Space set to ProPhoto RGB with no warning flags set on Color Space mismatch. With those settings, all images are brought into Photoshop and are automatically converted to ProPhoto RGB with no flags.

So, since I know of quite a number of Photoshop users that have those types of settings (including some members on ReTouch Pro and NAPP) this was not just a test yet an example how those folks can inadvertently end up with banding.

There are many other cases where visible degradation can occur and unfortunately, there is not a long listing of these cases anywhere for folks to refer to make sure they avoid. Editing in 16 bit mode dramatically avoids the degradation. So with that one step, you do not have to remember a long list of gotchas. That is more of my point.

Good question Kav. Please keep adding more comments as I post more cases

I will be posting another example shortly of another case.

[John Wheeler]

Degradation Case #2

I will leverage the same Sunset image from my previous post for this next case where you get degradation if you edit an 8 bit image in 8 bit mode instead of converting to 16 bit mode.

In this case:
- Open 8 bit sRGB image into Photoshop and keep sRGB as the same working space
- Convert to Lab mode (still in 8 bit mode)
- Duplicate two layers and set blend mode to multiply
- Flatten
- Convert to Profile back to sRGB

Here is a 100% view around the sun of the image. Notice that there is some banding. The banding is visible in prints.

Sunset 8 bit sRGB rountrip to Lab two multiply layers.jpg

Next image was created with these steps:
- Open 8 bit sRGB image into Photoshop and keep sRGB as the same working space
- Convert to 16 bit mode
- Convert to Lab mode (still in 8 bit mode)
- Duplicate two layers and set blend mode to multiply
- Flatten
- Convert to Profile back to sRGB
- Convert back to 8 bit mode

Here is the resulting image going roundtrip to Lab in 16 bit mode. Note that there is less banding:

Sunset 16 bit sRGB rountrip to Lab two multiply layers.jpg

The same conditions I mentioned from my prior post need to occur to see this banding e.g. low noise image and turn off dithering, viewing at 100%.

If your images either are not low noise to begin with or you leave dithering on, you probably will not see the banding which is great as long as you are fine with the higher noise.

I presented the above case because there have been statements that generically, you don't see degradation in 8 bit mode going back and forth to Lab. That may be a good general statement yet in my practical experience, I do see images that will visibly show degradation with 8 bit roundtrip to LAB. Similar to the last post, you are getting a little posterization going to a much wider Gamut Color Space which is amplified by some simple blends.

I will post some more examples yet may run out of steam before too long so the next post may be later (when there is daylight)

[kav]

I get the examples. It just seemed like only a proof of concept rather than a realistic workflow situation. I won't go further in depth on this as I don't want to derail a constructive thread onto the topic of color profiles.

[andrewrodney]

About the only time one may be stuck converting sRGB into ProPhoto is if you are building a composite image who’s parts are in ProPhoto RGB. Someone supplies you data in sRGB and that’s all they have. Its akin to those who unfortunately are handed a CMYK document and have to composite it into an RGB document. When someone hands you lemons, you try to make lemonade.

[Repairman]

Nice demo John. It provides a compelling reason to work in 16bit. Applying the same test to two images images where grads do not feature significantly, would we still notice a difference in quality. (I mean in commercial terms rather than 'in theory').
R.

[John Wheeler]

Originally Posted by kav

I get the examples. It just seemed like only a proof of concept rather than a realistic workflow situation. I won't go further in depth on this as I don't want to derail a constructive thread onto the topic of color profiles.

Hi Kav
Thanks for your additional comments.
Let me add a change to the last example to make it more realistic and a recommended and documented workflow (at least by some).

Before that let me say that I actually do not recommend making many of these adjustments in LAB and personally recommend using ACR or Lightroom even for adjustments on JPG and TIF files as the front end process. Those tools are very intuitive, non destructive, is done in high bit, and you can bring them into Photoshop as a Smart Object so you can have re-dos.
I also know that there are large number of Photoshop users that do not use Lightroom or ACR and continue to edit 8 bit images in 8 bit mode. I am just trying to put more information out there through examples so that readers/Photoshop users have a little more info of what lurks from within to make better decisions on their own workflow.

Degradation Case #3

This will be the same as my last post with simply replacing the two duplicated layers with multiply blend with a simple Levels Adjustments. This is a workflow that has been recommended by those steeped in LAB as a way to adjust luminosity without changing Hue/Sat.

For first iamge
- Open 8 bit sRGB image into Photoshop and keep sRGB as the same working space
- Convert to Lab mode (still in 8 bit mode)
- Add Levels Adjustment Layer
- Flatten
- Convert to Profile back to sRGB

Here is a 100% view around the sun of the image. Notice that there is some banding. The banding is visible in prints.

Sunset 8 bit sRGB rountrip to Lab One Levels Adj.jpg

Next image was created with these steps:
- Open 8 bit sRGB image into Photoshop and keep sRGB as the same working space
- Convert to 16 bit mode
- Convert to Lab mode (still in 8 bit mode)
- Add Levels Adjustment Layer with same settings as previous image
- Flatten
- Convert to Profile back to sRGB
- Convert back to 8 bit mode

Here is the resulting image going roundtrip to Lab in 16 bit mode. Note that there is less banding:

Sunset 16 bit sRGB rountrip to Lab One Levels Adj.jpg

The same conditions I mentioned from my prior post need to occur to see this banding e.g. low noise image and turn off dithering, viewing at 100%.

What was very interesting (and can be educational) about using this particular experiment in 8 bit mode is that as you move the sliders in the Lab Levels Adjustment Layer the banding streaks across the image like a psychedelic show. Sometimes single bands and sparse and sometimes a whole bunch. If you are not convinced there is an issue, this is the experiment for you.

[John Wheeler]

Originally Posted by Repairman

Nice demo John. It provides a compelling reason to work in 16bit. Applying the same test to two images images where grads do not feature significantly, would we still notice a difference in quality. (I mean in commercial terms rather than 'in theory').
R.

Hi R and thanks for your comments and thoughts. Good question about does it make a difference on a commercial basis and for images without slow low noise gradients. First let me say that visible issues can also occur in other situations as well. The most noticeable is losing details in shadows or highlights depending on how the processing was done. More on that later in additional posts when I can get to it.

Here is an analogy: Is it OK to drive without wearing a seat-belt? The vast majority of the time it is not an issue and it is a hassle to buckle up every time (try buckling in 3 very young grandchildren into car seats and you will understand the hassle ).

I would say that for many images there is not an issue. This is very similar to that many times working with JPGs is not an issue and just working with 8 bit JPGs out of your camera instead of RAW works most of the time too.

To me it is a time saver. I have been bit when editing 8 bit images in 8 bit mode several times and had to do rework. It was just a question was I willing to just go to 16 bit mode and double file sizes in general or just live with the occasional rework. I made the move to 16 bit for most cases and did not look back - no regrets for me. Yes there are cases where I just stay in 8 bit mode when I know if there were some subtle quality issues that it did not matter. When the quality mattered and I did not want to risk rework, I just stick with 16 bit.

If you have done lots of work and have never seen an issue, that is a good indicator that the type of images that you take and the workflow you use does not introduce a problem so your risks are pretty low of having an issue. The client may not notice or care if you had slight banding or loss of detail in shadows or highlights as well. So it is image, workflow, and client dependent whether it should be a concern for you or any other member.

I personally like the idea that I don't have to scour my images in every corner looking for these "8 bit image processed in 8 bit mode anomalies".

Such degradation is there in every 8 bit image processed in 8 bit mode, yet between the human eye and already existing or Photoshop introduced noise one does not see it most of the time - just like high quality JPEG compression. It is usually not an issue until you come across one you see and you say "what the @#$% is that?" and then move on to rework.

Thank you for you comments because I am hoping to move the discussion from "there is no degradation" to "yes there is degradation yet does it become visible with ones images and workflow often enough to be considered an issue that suggests staying in 16 bit would be a good idea." For me it did yet your mileage may vary.

[John Wheeler]

Degradation Case 4

Multiply Blend is "Extreme?"

It has been said on occasion that as long as you don't do anything "extreme", then processing 8 bit images in 8 bit mode is just fine. Of course the definition of extreme is not written down anywhere and usually it comes up when you have an image anomaly processing 8 bit images in 8 bit mode and someone helpfully tells you that you must have done something "extreme."

This degradation case is simply a mulitply blend followed by a curves adjustment Layer showing degradation in 8 bit that is dramatically less in 16 bit mode - I guess I am being "extreme"

Here is the base image at reduced resolution that I obtained from Stock Exhange: http://www.sxc.hu/photo/1138961 This image is in Adobe RGB and all I did for the starting point Convert to Profile...sRGB for consistency with the other cases.

Here is a reduced resolution version of the original.

Attachment 87808

Please notice the clocks in the shadows in the lower left corner on the clock tower.

A common technique for darkening an image and getting a certain effect is just to duplicate the image layer and apply a multiply blend. Here is that image:

Attachment 87809

Note that the clocks that were in the shadow are in the dark. I am just going to use a curves Adjustment Layer to recover those shadows in the following steps.

First here is a 100% of the original image. I did brighten it up by 2X to make it easier to see by using a Curves Adjustment Layer with white point at 128. You can see that the shadows have details:

Multiply is Extreme clocks 100% original 2X brightened.jpg

Here are the steps to create the following image.
- Convert to 16 bit mode
- Duplicate Image Layer and set Blend to Multiply
- Use Curves Adjustment Layer so Shadows to pull back the image similar to original.
- Brighten by 2X with another curves adjustment Layer for display on ReTouch Pro for better visibility (this extra step does not create degradation).
- Convert back to 8 bit mode

I targeted the clockface to have the same luminosity as the original. There is slightly higher tonal range due to the multiply:

Multiply is Extreme clocks 100% 16 bit Multiply plus curves 2X brightened.jpg

Now I repeated the steps in 8 bit mode
- No Conversion of bit mode it was already in 8 bit
- Duplicate Image Layer and set Blend to Multiply
- Use Curves Adjustment Layer so Shadows to pull back the image similar to original (same settings as previous image)
- Brighten by 2X with another curves adjustment Layer for display on ReTouch Pro for better visibility (this extra step does not create degradation):

Multiply is Extreme clocks 100% 8 bit Multiply plus curves 2X brightened.jpg

Now that's an image that only a parent or grandparent could love

So why such a large degradation for an 8 bit image in 8 bit mode processing. Turns out that Multiply/Screen blend is quite extreme blend in the shadows/highlights respectively.

For Mulitply Blend in 8bit mode all color pixel values at or below 37 out of 256 (~15% of your tonality) will be compressed down to the levels of 0 thru 5. So this causes posterization in the shadows of about 7X. When reversing course and trying to recover the shadows, the posterization remains.

In 16 bit Multiply Blend, that same tonality range up to 37 is compressed down to ~1400 levels (in 16 bit increments) due to the finer resolution of 16 bit mode. When recovering the shadows, virtually all the detail is still there.

Are there ways to avoid this in 8 bit mode. YES. Is the average Photoshop User aware of what to do - Based on my polling - NO. Basically, its best to avoid the pushing of pixels darker/lighter and then pushing them again lighter/darker respectively.

Now, who among us actually understands what is "Extreme" that will cause issues for 8 bit images in 8 bit mode with all the various Adjustment Layers, Blends, Layer Styles and their combinations among them against different image types. My background in Photoshop is from a forensics standpoint to understand when an operation actually reveals something valuable vs just introducing an anomaly. Even with 5 years of studying the details of the Photoshop operations from this standpoint I will not even make that claim of a complete understanding of what causes "Extreme." All I know is that I keep discovering more cases all the time.

Now you may ask "Are there any more types of anomalies in 8 bit processing of 8 bit images. YES. The most common is increased additional image noise. Every operation in Photoshop has what is called rounding errors with each operation creating a little more noise with each operation. Some of the noise cancels each other out and some does not. If each pixel experiences just a few operations, you would most likely not see the issue. On the other hand, if each pixel experiences 100 operations (heavy duty Layer Stack, blends etc) then it can be noticeable.

In 8 bit mode an error on the order of +/- (1/512) is introduced with each operation. In 16 bit the error introduced with each operation is +/- (1/131,072). A much smaller error to be accumulating with less introduction of this rounding noise.

I do not time to post another example showing this noise at this point and if there is strong interest I could do so yet wouldn't be until after next week.

So should you process 8 bit images in 16 bit mode? Hopefully this and the previous posts will provide some practical examples to help you make a decision that works best for you and your workflow.

Thanks for putting up with my long posts. I hope this was useful to someone (besides to read and help you go to sleep at night zzzzzzzzz )

[Tony W]

It has taken me quite a long time to reply to this thread (due to personal reasons causing me to have limited time). Please accept my apologies particularly John as you obviously went to some trouble and spent a lot of time with your response. I have to say that I have not actually repeated your efforts although I did not follow exactly your procedure

I have only gone so far with this looking at the first case.

Degradation case 1
The original image without adjustment shows obvious signs of posterisation therefore normally care should be taken in handling.

Using the image you posted from photobucket followed I think your example with one exception I did not go into ProPhoto RGB for the reason that banding could be worsened.

I normally leave Dither on but in this case turned off to better mimic your tests:
Steps taken for my image:

• Saved image as is from Photobucket flagged as sRGB
• Opened in Photoshop as original 8 bit sRGB
• Duplicated and converted this to 16 bit
• 8 Bit background layer duplicated and set blend to Screen
• Dragged the duplicate layer down to copy. Ending up with 3 layers background, layer 1 and 2 both set to screen blend
• Flattened image
• Edit > Convet to sRGB profile ( I know I was already there but wanted to copy your workflow)
• Followed the above for the 16 bit conversion.

My attachments show what I got - but I have to say that the save for web conversion seems to have thrown something into the mix as these images matched exactly on screen. I have not tried printing yet but I think you may agree that if there is a difference it is very difficult to see.

In fact in positioning the second image I was unsure which was which (should have named the layers better)

I have not tried your second or third degradation examples due to time constraints and the fact that LAB has been thrown into the frame.

Generally I am happy to work in LAB where I think the image would benefit. In this case however I can think of no benefit or reason to turn to LAB, I would not want to brighten or separate the colours for instance.

There has been much discussion/controversy about LAB conversion FWIW I have not noticed so far any losses worth worrying about on the images I have tackled, but I do not support the view that the image has not lost anything – I understand that 20-30 levels may be lost in the first conversion – so something has gone but in some cases this may not be missed, for others it may be important?

I personally like the idea that I don't have to scour my images in every corner looking for these "8 bit image processed in 8 bit mode anomalies".

I can appreciate this viewpoint

[John Wheeler]

Hi Tony
Sorry but your post made me smile a bit. Here is a quote from your post:

Originally Posted by Tony W

......Degradation case 1
Using the image you posted from photobucket followed I think your example with one exception I did not go into ProPhoto RGB for the reason that banding could be worsened.

Yet in my post for Degradation Case 1:

Originally Posted by John Wheeler

Degradation Case 1
So here is a case where you will see banding degradation when an 8 bit image is processed in 8 bit mode yet not when it is processed in 16 bit mode. This occurs when this 8 bit sRGB image is converted to a working space of ProPhoto RGB.

So of course, if my case is to show the degradation differences between 8 bit mode and 16 bit mode when going to ProPhoto RGB, and you eliminate the step of going to ProPhoto RGB space (the whole point for this degradation case), of course you will not see the same result

So I apologize if I missed your point, yet what is your point?

Just to summarize what I was trying to indicate. There has been long standing discussions that converting 8 bit images to 16 bit makes no difference in image quality. I would agree with that until ones start to do additional Photoshop processing. With additional processing, there can be a visible difference if taking an 8 bit image and staying in 8 bit mode for additional processing vs taking a 8 bit image and converting to 16 bit before doing additional processing. I presented just 4 degradation cases (and there are many more) where there is visible degradation. Here is the combined summary of those 4 cases:

Case 1: Taking an sRGB image and processing in ProPhoto RGB space and back to sRGB. There are many Photoshop users that have their Color Settings set to a default color space of ProPhoto RGB or take JPEG images through Lightroom or ACR into Photoshop as ProPhoto RGB with no flags set to let you know you are converting from sRGB to a wider gamut space. Then when outputting to Web or for printing it is converted to sRGB again. Not all images will show degradation yet I documented the types of images that very likely would.

Case 2/Case 3: This was moving an sRGB image to LAB for special processing and then back to RGB. Repairman indicated that Case 2 may not be common ("in theory") so I included Case 3 following steps right out of Lab books. Again, not every image will have visible degradation yet again, I indicated the types of images where where degradation would be visible.

Case 4: This case showed degradation while staying in just sRGB and not changing color spaces using a simple multiply blend and Curves Adjustment Layer. I have seen these steps documented in many Photoshop books so assume many Photoshop users may follow these same steps. Not all images will show visible degradation yet again, I documented the cases in which the degradation would be visible.

There is a myth that moving 8 bit images to 16 bit mode has no visible quality advantages for normal additional Photoshop processing. For many images that is true. To make that a blanket statement for all images is just that - a myth and is simply not true (IMHO ). So should everyone just use 16 bit - NO. There are tradeoffs with file size, processing speed, and available functionality in Photoshop. Just don't depend on that myth to cover your rear and assume that it never matters.

Now as a side note - Tony, the images that you were allowed to download from my Photobucket account were apparently not full resolution. I don't know why my account did not allow a full resolution download for you. I checked and the lower resolution images for Degradation Case 1 worked the same at lower resolution as well (if you follow all the steps ). For anyone wanting the full resolution images for any of the cases just send me a PM and I will figure a way to send them through either email or another posting mechanism.

[Tony W]

Sorry but your post made me smile a bit.

No need to be sorry, always glad to bring a smile to someone’s face even when I do it inadvertently
So that I do not take anything out of context (hopefully) I will use your reply as my basis and offer explanation comment where appropriate

So of course, if my case is to show the degradation differences between 8 bit mode and 16 bit mode when going to ProPhoto RGB, and you eliminate the step of going to ProPhoto RGB space (the whole point for this degradation case), of course you will not see the same result

My turn to be sorry John but I totally missed the point and should have studied your reply more thoroughly before responding. In the event I saw that the move was in Prophoto and it looked like you had just decided to convert to this for no apparent reason other than your standard working space. If I had properly read your earlier statement you did make it clear that the degradation can occur with 8 bit sRGB is converted to ProPhoto RGB - a fact that I can attest to

So I apologize if I missed your point, yet what is your point?

If you did miss my point (although I do not think this to be the case) the blame can only be laid at my feet. To ask what is my point further strengthens the case that I have perhaps been a little vague in my first post.

Two reasons for this, first I was trying not to be too ‘wordy’ and keep my question short and concise. Secondly I was attempting not to be prescriptive as I was looking for other opinion which may or may not coincide with my own.

So the points I set out to try and clarify were:

• Any real benefit taking an 8 bit file and editing in 16 bit workspace and potential pitfalls. Some ‘gotchas’ I was already aware of and which you have highlighted in your replies.

• Is it possible that 8 bit to 16 bit could degrade image due to moving to the higher bit depth. This thought stemmed from a conversation with another PS user who was quite adamant that due to losing so many levels there were huge gaps in the histogram and these were not evident in the PS histogram views. Further he maintained that these gaps could be ‘repaired’ with a plug in (name unknown) he used, therefore avoiding potential image degradation problems. My first thoughts were that he had bought into buying ‘snake oil’. How could a histogram be repaired other than perhaps to interpolate the missing information in the gaps and would this actually make an observable difference? Perhaps these thoughts were unfair and incorrect and the application may offer something?

Case 1: Taking an sRGB image and processing in ProPhoto RGB space and back to sRGB. There are many Photoshop users that have their Color Settings set to a default color space of ProPhoto RGB or take JPEG images through Lightroom or ACR into Photoshop as ProPhoto RGB with no flags set to let you know you are converting from sRGB to a wider gamut space. Then when outputting to Web or for printing it is converted to sRGB again. Not all images will show degradation yet I documented the types of images that very likely would.

My default colour space is ProPhoto RGB but all flags are set to warn of conversion mismatch. But can understand the warning for the unwary.

Case 2/Case 3: This was moving an sRGB image to LAB for special processing and then back to RGB. Repairman indicated that Case 2 may not be common ("in theory") so I included Case 3 following steps right out of Lab books. Again, not every image will have visible degradation yet again, I indicated the types of images where where degradation would be visible.

Not really had time to study these but I am sure will be of interest.

Case 4: This case showed degradation while staying in just sRGB and not changing color spaces using a simple multiply blend and Curves Adjustment Layer. I have seen these steps documented in many Photoshop books so assume many Photoshop users may follow these same steps. Not all images will show visible degradation yet again, I documented the cases in which the degradation would be visible.

Again need to have a proper look at this. Multiply blend is something that I rarely need to use these days but I think I see where you are coming from

There is a myth that moving 8 bit images to 16 bit mode has no visible quality advantages for normal additional Photoshop processing. For many images that is true. To make that a blanket statement for all images is just that - a myth and is simply not true (IMHO). So should everyone just use 16 bit - NO. There are tradeoffs with file size, processing speed, and available functionality in Photoshop. Just don't depend on that myth to cover your rear and assume that it never matters.

I can appreciate your points here and to make generalised statement is wrong without pointing out where these generalisations may fall apart. I can also understand that it may be worthwhile in switching to 16 bit to attempt to cover your rear. The tradeoffs with file size and processing speed are not a huge concern to me at least. Functionality loss may be problematical at times and then there is the fact that it is likely that you will need to output at some time back in 8 bit.

Now as a side note - Tony, the images that you were allowed to download from my Photobucket account were apparently not full resolution.

No problem John I do not think it too important as I can see where you are coming from and me taking the same actions will result in what you have demonstrated.

[John Wheeler]

Originally Posted by Tony W

Two reasons for this, first I was trying not to be too ‘wordy’ and keep my question short and concise.

Unfortunately my replies are wordy. An affliction of which I have not cured myself.

Secondly I was attempting not to be prescriptive as I was looking for other opinion which may or may not coincide with my own.

As per our PMs even from the previous thread I appreciate you posting this thread and never took your words as prescriptive. My responses were to let some real visible examples do a little MythBusting rather than just supply opinion.

So the points I set out to try and clarify were:

• Any real benefit taking an 8 bit file and editing in 16 bit workspace and potential pitfalls. Some ‘gotchas’ I was already aware of and which you have highlighted in your replies.

I think that one was beat to death for this thread (I think). Visible degradation in some images in some common circumstances and not visible in many more images (IMHO)

• Is it possible that 8 bit to 16 bit could degrade image due to moving to the higher bit depth. This thought stemmed from a conversation with another PS user who was quite adamant that due to losing so many levels there were huge gaps in the histogram and these were not evident in the PS histogram views.

Yes and No

The conversation with the other PS user sounds like more of a misunderstanding of 16bit mode and associated 16 bit math used by PS. Hard to know without participation by that person. I personally don't know any case where using a higher precision data and math (16 bit mode) in Photoshop creates a worse result. I will leave that to someone posting an example.

That said, there is a case of cumulative degradation (added noise) when making round trips to 16 bit mode (not 16 bit mode created yet from the conversion). That is when the Dither Mode is check on in Color Settings. Every time you return from 16 bit mode back into 8 bit mode with Dither turned on, Photoshop adds random dither noise into 0.5% of your image pixels (Dither being 1 bit random variation per color channel). I confirmed this with an Adobe engineer and a documentation bug was logged. That noise is very difficult to see even with slowly changing gradients so is mosly an academic example. It would require many round trips to 16 bit and back for this noise to accumulate and be noticeable and who does that?

The tradeoffs with file size and processing speed are not a huge concern to me at least. .

For many it may not be an issue. The way I look at it is when the combination of File Size, number of Layers, number of History States are large enough you eventually run out of allocated PS memory and PS ends up using your scratch drive extensively with a large performance slowdown. Basically its the performance "wall." Running in 8 bit mode pushes out that wall by 2X in base pixel image size. If you are within 2X of that wall in 8 bit mode and switch to 16 bit mode, you will hit it. Happens to me all the time yet I often work with very large files.

Thanks again Tony for posting this thread.