Silverfast 8: 16bit grey scale?

[Adrian Bacon]

After some further testing...

Unaltered Vuescan generated files. Acquire 8 bits, save 8 bits raw, Adobe RGB tone curve, then acquire 8 bits, save 16 bits raw, then acquire 16 bits, save 16 bits raw. The file names match the the type.

http://m.avcdn.com/sfl/vuescan-08-vs-16-step-wedges.zip

The histogram for the acquire 8 bit save 8 bit raw

Now the acquire 8 bit, save 16 bit raw

Now the acquire 16 bit, save 16 bit raw

The two 16 bit raw saves are a very close match, and very different than the 8 bit acquire save 8 bit raw, so they should be able to be compared to each other for the purposes of looking at dynamic range differences. I can generate and upload just some black frame noise floor images if anybody wants to look at those as well, though toggling between the two 16 bit raw save histograms shows that there is a difference between the two down on the bottom end of the scale.

I've not looked super close at either 16 bit raw save as of yet to try to quantify any differences between the two other than noting that the histograms have differences down on the bottom end.

 

[Adrian Bacon]

Doing a little more testing, and it appears that acquiring single channel monochrome 8 and 16 bit raw also has the same behavior as full color, so as long as you're saving 16 bit raw monochrome, you can acquire 8 bit or 16 bit and both will be saved with no gamma applied.

I did a quick check of single channel black frame of both 16 bit and 8 bit acquire with the exposure boosted in post so its easier to see the noise differences between the two. Both files have had the same operations performed on them.

acquire 8 bits, save 16 bit raw, histogram

acquire 16 bits, save 16 bit raw, histogram

The width of the histogram is a reflection of the amount of noise present. The cleaner the noise floor is, the skinnier the spike in the histogram is.

Zooming in and looking also reflects that acquiring in 8 bits is a bit noisier as well.

acquire 8 bits, save 16 bit raw

acquire 16 bits, save 16 bit raw

And finally, the raw Vuescan files so others can look and draw their own conclusions: http://m.avcdn.com/sfl/vuescan-08-vs16-black-frames.zip

It should be noted that in actual images of BW negatives, these differences will be masked somewhat due to the film grain as @alanrockwood has been exploring with a number of his posts, and the noise differences will largely show up in the highlights (after inversion), etc, and whether any of this makes any difference really depends on what you're going to do with the files afterwards, but I think it's clear that acquiring in 16 bit mode is a bit cleaner than acquiring in 8 bit mode, and unless doing so is just not possible, acquiring in 16 bit mode is generally preferable over acquiring in 8 bit mode. Whether you save in 8 bits or 16 bits I think comes down to whether you're going to do your post processing afterwards in another program, or are doing it in the scanner software.

Another observation that I've not yet been able to explain, but has me wondering if the 8 bit acquire really is 8 bit acquire is the number of unique tone values when saving 16 bit raw from 8 bit acquire. If we really are only getting 256 unique values from the scanner in 8 bit acquire, we shouldn't have more than 256 unique values in the resulting saved 16 bit file, but that doesn't seem to be the case when acquiring in 8 bits and saving in 16 bit raw. We should also have the minimum non-zero value be no less than 257 after scaling to 16 bits for save, but that also is not the case, it's much less than that. I've not yet quantified how many unique values there actually are, but just at first blush, it strikes me as having quite a few more than that. I don't think it's a full 16 bits, but I wouldn't be surprised if it's more along the lines of 10 or 12 bits coming off the wire, or some other internal firmware downscaling from a higher bit depth to 10 or 12 bits. More investigation needed there.

EDIT:

The minimum non zero value of the 8 bit acquire is 16, and the minimum non zero value of the 16 bit acquire is 1, which would seem to suggest that in 8 bit acquire, it's really 12 bits being acquired (1 / 4095 * 65535 = 16) that is then scaled up to 16 bits, which is also probably why Vuescan always gamma encodes 8 bit saves, no matter the acquire bit depth, at least with Epson scanners. It does it because it's getting more than 8 bits from the scanner in 8 bit mode.