Converting 8-bit to 16-bit

[sanking]

16-bit files are much less prone to posterization than 8-bit files if one does a lot of tonal correction on the file in Photoshop. A while back I asked for suggestions on how to deal with some 8-bit drum scans of 12X20" negatives which I did not want to re-scan. I received some good adivice on this forum and have already applied it to several of the scans.

There is also another issue. Suppose you have a lot of existing 8-bit RGB files, or own a drum scanner or high-end flatbed that only scans in 8-bit. I asked this question on the ScanHi-End Group on Yahoo, and received a number of suggestions for converting 8-bit to 16-bit. The system that has worked best for me in scanning LF B&W negatives is this.

1. Scan in 8-bit RGB, doing as many tonal corrections as possible *before the scan." The reason for this is that many drum and high-end flatbed actually scan in 14-bit or 16-bit, and then convert to 8-bit for save. If you do as many corrections as possible before the scan these are done during the scan in either 14-bit or 16-bit.

2. After the scan, open the file in Photoshop and then before doing anything else, convert the file to 16-bit RGB. If you are working with a LF negative this is going to give you a huge file, so be patient.

3. Next, change the file from 16-bit RGB to 16-bit Grayscale, then save. You can do further corrections on this file as if it had been originally scanned in 16-bit.

If you look at the histogram after converting to 16-bit grayscale you will see that it is much smoother than an 8-bit grayscale file, with just a few spikes. Not as smooth as a file scanned originally in 16-bit, but not far off, and so far I have not seen any posterization with this method.

This method could also be extended to any existing 8-bit RGB files.

Sandy King

[vinny]

Thanks Sandy, i've been wondering about the difference in my large format scans myself. I made a scan from the imacon in 16bit and in 8bit. I could see a bit of difference but not much on my screen. I bet it would be more apparent on the print. Interesting discovery on the posterization.

vinny

[PViapiano]

There was info re this just a few months ago in the forum, in fact, everyone jumped all over a guy that was proposing 8 to 16 bit conversions.

The final word, though, was, once a file is in 8 bit, a conversion to 16 bit does not buy you anything. I understand what you're saying about 8 bit scanners actually doing their scans in 14 or 16 bit, and I can visualize that that could help you get a more robust file at scan time theoretically, but once it's knocked down to 8 bit, you cannot recover any of those lost bits.

I'm sure many more will chime in here...

[Jiri Vasina]

You don't recover any lost bits this way. But any further processing of 16-bit image should result in smoother transitions, there are more steps you can fill. So it's less likely you'll introduce any digital artefacts. If you don't do any curves/levels/burn-and dodging treatment or whatever, you don't gain a thing. But the more of these manipulations you do, the better to have the file in 16bit... (or at least that is the way I rationalize it)

[Ted Harris]

vinny,

The differences in a full range color print are subtle but immediately apparent. The Canon 5000 export driver processes images in 16 bit and sends them to the printer as 14 bit. I printed an image using the export driver and the "print w/preview" from PS CS2 and have now had dozens of folks look at the two prints. The 16 bit is the winner everytime.

[Ted Harris]

Sandy, that is interesting! Have you found any documentation taht verifies that your Eversmart Pro is actually working in the 16 bit mode when making prescan adjustments?

[Jack Flesher]

Thanks for the confirmation Sandy! I was one of the folks (or the one) who said there was benefit to converting from 8 to 16 before doing any processing as it gives you more adjustment range, and yes, I got jumped on.

But the fact is I had tested it out for myself long ago and confirmed that I got a better final histo by doing the conversion. And regardless of the naysayers, it does make logical sense -- again, you are not creating any new original tones, but are allowing finer gradations during any image manipulations after the conversion, which then helps resist banding in even-toned areas like skies as one main example.

Cheers,

[sanking]

Sandy, that is interesting! Have you found any documentation taht verifies that your Eversmart Pro is actually working in the 16 bit mode when making prescan adjustments?


Ted,

What I understand from reading the Scitex literature is that after the prescan adjustments the actual scan is made in 14-bit, which is then processed to 16-bit, and then compressed to 8 bit.

Quote from a Scitex EverSmart/EverSmart Pro "Product Booklet".

"The Scitex EverSmart scanner works with one Tri-linear CCDE, 8000 pixels with a built-in Anti-Blloming feature. The analog information is converted into 14 bit/color/pixel by A/D convertor and then compressed to 8 bit/color pixel."


Sandy King

[Larry H-L]

Hmm... I recently received some 8-bit Tango drum scans from 5x7 negs. When I converted them to 16-bit, I immediately saw banding in the sky areas. Not sure if it was real or a screen issue, but the banding went away when I converted back to 8-bit. Any thoughts?

[Brian Ellis]

vinny,

The differences in a full range color print are subtle but immediately apparent. The Canon 5000 export driver processes images in 16 bit and sends them to the printer as 14 bit. I printed an image using the export driver and the "print w/preview" from PS CS2 and have now had dozens of folks look at the two prints. The 16 bit is the winner everytime.

Ted, sorry if this is a dumb question but I don't understand what you're saying here. Are you saying that with two otherwise identical prints, one printed from a file sent to the printer as a 16 bit file, and the other as an 8 bit file, the 16 bit file produces a noticeably better print? In other words, if I do everything in 16 bit up to the point of sending the file to the printer and at that point make a copy of the file and convert it to 8 bit, then send both files to the printer (i.e. one 8 bit file, one 16 bit file), there will be a noticeable difference in the two prints? Or are you just saying that there's an advantage to doing what Sandy is talking about and converting an 8 bit file to a 16 bit file before editing it?

[PViapiano]

Jack, I don't recall you as the guy everyone jumped on. Here's the thread I recall:

http://www.largeformatphotography.info/forum/showthread.php?t=20558&highlight=scanning+tutorial

In the end, you have to do your own experimentation and see what looks OK for you, same as film testing, etc, blah, blah...

Nothin' from nothin' leaves nothin'...

[Jack Flesher]

Jack, I don't recall you as the guy everyone jumped on. Here's the thread I recall:

http://www.largeformatphotography.info/forum/showthread.php?t=20558&highlight=scanning+tutorial

In the end, you have to do your own experimentation and see what looks OK for you, same as film testing, etc, blah, blah...

Nothin' from nothin' leaves nothin'...

Ah, sorry... I was referring to the thread a month or so ago where Sandy asked the question originally...

IMO it is more akin to getting "something from nothing" but respect YMMV ;)

Cheers,

[sanking]

Thanks for the confirmation Sandy! I was one of the folks (or the one) who said there was benefit to converting from 8 to 16 before doing any processing as it gives you more adjustment range, and yes, I got jumped on.

But the fact is I had tested it out for myself long ago and confirmed that I got a better final histo by doing the conversion. And regardless of the naysayers, it does make logical sense -- again, you are not creating any new original tones, but are allowing finer gradations during any image manipulations after the conversion, which then helps resist banding in even-toned areas like skies as one main example.

Cheers,

Well, the method I mentioned was suggested by someone on the ScanHi-End Yahoo group. Many of those folks scan for a living and so I take their advice seriously. You can have a look at the discussion by entering the group here. http://tech.groups.yahoo.com/group/ScanHi-End/message/15639

There is no question but that this method significantly reduces the risk of posterization compared to making the original scan in 8-bit grayscale and then making tonal corrections in Photoshop in 8-bit format, or for that matter, the result is better than scanning in 8-bit grayscale, converting to 16 bit gray, and then making tonal corrections. Again, the method calls for making as many tonal corrections as possible in the prescan adjustments, scan in RGB, open in Photoshop and convert the 8 bit RGB file to 16 bit RGB, and then finish by converting the 16 bit RGB to 16 bit grayscale. You are not creating more tones but subsequent tonal corrections are much less likely to cause posterization. This is obvious, when you do a comparison, in the spikes of the histogram after adjustments, and in the final print.

Clearly the best method is to scan and save in 16 bit if that is available. But this method is a good work around if you happen to own an older drum scanner or high-end flatbed that does not save files in 16 bit.

Sandy King

[Kirk Gittings]

I have noticed the same possibilities as I sometimes use an 8 Bit drum scanner in Santa Fe. This works with good drum scans because the noise level is so low. I don't find it as workable with prosumer flatbeds.

[Ted Harris]

Brian,

No I am not saying the same thing Sandy is saying. What I am saying here is that, assumin gyou start with a 16 bit image and do all of your processing in PS in 16 bit and then send it to the printer (in this case a Canon ipF 5000-5100-6100-8000-9000) as a 16 bit file any further processing you do in the Canondriver before printing will be done in 16 bit and then sent to the printer as 14 bit. OTOH the Epson printer driver that you access through the PS "Print with Preview" window (or Canon if you go that way), is only an 8 bit driver .... Thus, what I did was print the same file as an 8 bit file sending it to the printer through the "Print with Preview" window from PS to the Canon 5000 and sending it to the printer as a 16/14 bit file through Canon's Export driver. I laid them out side-by-side and asked people which they preferred. Everyone picks the 16 bit print but the differences are so subtle that none of the lay viewers and for that matter only one photographer who is also a master digital printer can actually point to any parts of the print where they see or think they see differences.

[sparq]

Hmm... I recently received some 8-bit Tango drum scans from 5x7 negs. When I converted them to 16-bit, I immediately saw banding in the sky areas. Not sure if it was real or a screen issue, but the banding went away when I converted back to 8-bit. Any thoughts?

Displaying of 16-bit gray scale images may have its issues on stock hardware. Think of dithering etc.

[Henry Ambrose]

Ted,

What I understand from reading the Scitex literature is that after the prescan adjustments the actual scan is made in 14-bit, which is then processed to 16-bit, and then compressed to 8 bit.

Quote from a Scitex EverSmart/EverSmart Pro "Product Booklet".

"The Scitex EverSmart scanner works with one Tri-linear CCDE, 8000 pixels with a built-in Anti-Blloming feature. The analog information is converted into 14 bit/color/pixel by A/D convertor and then compressed to 8 bit/color pixel."


Sandy King

About that quote from your Scitex booklet -
(for friendly conversation and not just picking a nit)

I don't read this as verifying the internal corrections being made in 16 bit. What I see is that the AD (analog to digital) convertor outputs 14 bit which is then compressed to 8 bit for output. So the corrections may be made on 14 bit info but we don't know that for sure. They may just as well be made during the 14 to 8 bit conversion or after. I don't know. It'd be interesting to know for sure how it works. But I read nothing that indicates the info is ever 16 bit.

And about the appearance of histograms - Just because they are smooth doesn't mean they are good. A bumpy histogram may simply indicate that the file contains "bumpy" values which came from "bumpy" film densities which match the scene information.

[sanking]

About that quote from your Scitex booklet -
(for friendly conversation and not just picking a nit)

I don't read this as verifying the internal corrections being made in 16 bit. What I see is that the AD (analog to digital) convertor outputs 14 bit which is then compressed to 8 bit for output. So the corrections may be made on 14 bit info but we don't know that for sure. They may just as well be made during the 14 to 8 bit conversion or after. I don't know. It'd be interesting to know for sure how it works. But I read nothing that indicates the info is ever 16 bit.

And about the appearance of histograms - Just because they are smooth doesn't mean they are good. A bumpy histogram may simply indicate that the file contains "bumpy" values which came from "bumpy" film densities which match the scene information.

You are right. I am not sure how this is done, but the Scitex booklet clearly indicates that the analog data is captured in 14 bit, then converted to 16 bit, and eventually compressed to 8 bit. And I was told by a former Scitex technician that image processing was done in 16 bit. What would be the point in capturing in 14 bit - 16 bit, and then compressing to 8 bit before image processing? That sounds like brain dead technology to me.

In any event, that is all pretty much irrelevant. Doing the capture and saving the way I described, with this scanner, gives me better histograms, and better tonal values on the print.

Sandy