What's the real bit depth of Epson V700?

[Rudeofus]

According to the Epson data sheet they produce 16 bits for each color channel, but is this really what the sensor produces in terms of actual information? Does anyone have actual numbers, or does anyone have a V700 scan (as 16 bit TIFF) of a very uniform grey area so I could do the math?

 

[pellicle]

Hi

what matters is ones ability to tune the analog sensors input levels to match your AD converter. This needs to be done on specific films and is something that can't be done on a flatbed. I'm guessing that the Epsons are tuned to put their white on the calibration reading and their black at their dmax

I've tested all my Epsons with a Stouffer wedge and seldom does the DMax go as dark as them.

you may find the following helpful reading, note that the 4990 (probably the closest to your 700) does not significantly penetrate into the deep black of the stouffer better than the 3200


http://cjeastwd.blogspot.com/2009/04/epson-4990-response-testing.html

http://cjeastwd.blogspot.com/2008/03/using-epson-flatbed-scanners-for.html

essentially its a non-question (and rather like asking if a Ferrari is faster than a Lambo) as only drum scanners have the ability I mentioned above and the older ones were often only 8 bits per channel anyway.

I think there is a good case for arguing that a well captured 8 bits provides plenty of leeway for adjustment without posterisation and the main pressing need for 16 bits is in hardware which can not be tuned (such as any flatbeds I can think of {anyone know if you can set hardware levels on the Creo?}).

I think that you will get more from learning to drive the V700 series well than fretting about specs of these. They are good tools for the money.

I await some scans back from a friend with a Flextight X1 to make objective comparisons to my 4990

stay tuned for that one

 

[Loris Medici]

pellicle, I don't understand this. Do you indeed say that a well captured 8 bit file provide enough room for adjustments just as a 16 bit file!? Not so in IME, especially in the context of b&w film; think of the levels and gamma adjustments you have to do just to linearize the raw scan. (= The absolute minimum manipulation.) That brings us to the natural conclusion that hardware 8 bit is pretty much a [POOR] figure too. (Drum scanners are different animals; they usually can see grain, and at that level you don't need much bit depth. Plus, they can do multiple passes by design, increasing the effective bit depth...)

Regards,
Loris.

...
I think there is a good case for arguing that a well captured 8 bits provides plenty of leeway for adjustment without posterization and the main pressing need for 16 bits is in hardware which can not be tuned (such as any flatbeds I can think of {anyone know if you can set hardware levels on the Creo?}).
...

 

[pellicle]

Loris

Do you indeed say that a well captured 8 bit file provide enough room for adjustments just as a 16 bit file!?

yes, that's what I said. Chances are however that unless you are tuning your hardware at scan time to set that analog gain of your photo-receptors to the black and white points of the media and then dividing that among the 8 bits per channel you are not going to be able to do it.

I have had a number of scans done which were returned to me as 8 bit tiff files, I have then done some adjustments on these files and subsequently converted to another profile, I saw no evidence in the prints I got to have "ahhah, posterisation banding" jump out at me.

If you are doing large adjustments I would argue you did not scan it perfectly.

It is important to keep in mind what I said and not take this to an extreme and remember I said well scanned.

Yes, I know we have been taught to perfer 16 bit, but after I've cleaned up an image and got it how I want I will often convert it to 8 bits (but only if I'm sure and I've flattened the image of any layers).

for instance I have scanned this image (not this file, but this image)

Dead Link Removed and after being satisfied with it have flattened it and then sent it off for printing. Didn't like the contrast, adjusted a wee bit and printed again to my satisfaction and can't tell the difference.

same with this one (scanned on a eversmart flatbed I suspect)
Dead Link Removed

printed, tweaked and reprinted (this time not to my satisfaction) but with no evidence of posterisation.

If you want I can roll up that last one and post it to you as I'm moving back to Australia and may as well give it away (its 77cm wide). Plop me a couple of euros postage and I'll whack it in the mail for you

 

[pellicle]

PS I am not recommending you scan in 8 bits, especially with most consumer type scanners. Nor am I saying that 8 bits is better to edit than 16 bits (though it can make RAM requirements lighter on big images).

PPS

if you examin arguments like on this page http://jtrujillo.net/digital-photo-tutorials/8vs16bit/

you will find their proof of 16 bit advantage in quite severe adjustments:

like he does this to the image:

then undoes it with:

whoa, I'm not likely to do such wild adjustment to any of my images ... no wonder he got banding

 

[Loris Medici]

It's understandable to convert finished/manipulated files to 8 bit, "ONLY for sending them to the lab for printing". I would always keep the 16bit master file.

For small manipulations (like your suggestion above: small/fine tweaking) 8 bit files could do the job indeed. BUT, I'm talking about raw scans from B&W film, 1) linearization and 2) manipulation:

1. Think of a pyro developed B&W negative; those aren't dense, you'll often have to do a pretty strong levels adjustment, even in case of PERFECT exposure. Then you'll have to linearize the tones which involves a pretty severe gamma correction.

2. Burn/dodge/local contrast adjustments with masks and what not with several layers. (That's how I work...) In that case you'll often notice a drop in quality if working with a 8 bit source. And sometimes you indeed have to do severe LOCALIZED adjustments, not every adjustment has to be applied to the whole file...

Sorry, I'm not convinced at all; I will continue to scan and work in 16bit mode and I definitely don't recommend anyone else to do otherwise. It's pointless unless you have a pretty poor system by todays standards.

Regards,
Loris.

 

[pellicle]

Loris

Sorry, I'm not convinced at all; I will continue to scan and work in 16bit mode and I definitely don't recommend anyone else to do otherwise.

I appreciate that english is not your native language but I was not trying to convince anyone of anything; to wit I actually stated:

I am not recommending you scan in 8 bits, especially with most consumer type scanners.

perhaps that was not clear enough ... I will repeat it again more clearly

I am not recommending you scan in 8 bits

I hope that clears up my situation

 

[Loris Medici]

I don't say that you recommend to scan in 8bit anywhere...

My objection is completely towards your original statement: "...I think there is a good case for arguing that a well captured 8 bits provides plenty of leeway for adjustment without posterisation..."

This could (actually will) lead unknowledgeable people to use 8bit files (in capture and post-processing) which is totally bad practice. Besides, frankly I think there's nothing to argue about it...

Regards,
Loris.

 

[sanking]

My Eversmart Pro makes the analog to digital conversion in 14 bit but the software only allows 8 bit saves. My work procedure for this scanner is to make as many corrections as possible in the pre-scan so that all of the very abrupt tonal transitions are done in the high bit scan. I then save the file as 8 bit, but immediately convert it to 16 bit before doing any final corrections or changes. Converting an 8 bit file to 16 bit will not add any new tones, but it will prevent losing any of the existing tones when making additional corrections. I get very good results with this workflow scanning LF negatives in grayscale.

BTW, with the Eversmart there is nothing to be gained by scanning a B&W negative (not stained) in RGB as there is virtually no difference in grain or sharpness in the three channels.


Sandy King

 

[pellicle]

it will be nice when we can make adjustments in floating point rather than integer maths ...

 

[pellicle]

Hi

BTW, with the Eversmart there is nothing to be gained by scanning a B&W negative (not stained) in RGB as there is virtually no difference in grain or sharpness in the three channels.

I just keep hearing good stuff about them ... if only they weren't so heavy to ship to Australia

sigh

 

[Doug Fisher]

>>>I await some scans back from a friend with a Flextight X1 to make objective comparisons to my 4990<<

I hope he set the software sharpening to -120 so that you get an apples to apples comparison that cancels out the default behind the scenes sharpening. The X1 produces nice scans - but it should considering the price

Doug

 

[pellicle]

Doug

nice point, I'll email him to ask.

I'll letcha know.

I really should do a blog article on your scanner ... its excellent

 

[Rudeofus]

pellicle,

can you help me interpret the results from your stouffer wedge link? Is it indeed the case that when scanning with 8 bits I can expect pixel values between 17 and 250 or so? Or did I miss something in this test?

 

[pellicle]

Hi

yes, but my procedure was to not alter the black and white end points in the driver, keeping everything fully wide. I did this because altering the data set in software will make for complex variables. I wanted what the unaltered fully white reported to and what the fully black reported to

if that is clear?

by setting levels in the dialog box of the Epson scan you take what the scanner gets and re-map it.

Interestingly I do not get 255 anywhere even on the supposedly uncovered fully exposed areas. this indicates that the calibration of hardware is not perfect.

I'm intending to remove the glass in the future to see if this gives me access to a better white level.

my full data set is available here. I have also done experiments where I cover the "calibration" area of the scanner with unexposed but developed film base and that is also in that data set. This is presented as plain, 1 sheet, 2 sheets; in the 3200 extras

The scan was done at 16 bit, however I used photoshop to measure the values by selecting an inner portion of a square and seeing the values of this in the information dialog which remain as 8 bit represntations (NB 0 to 255).

 

[Rudeofus]

Now that data set makes me really curious: At the darker sections you claim a standard deviation of 1.4 - 2 %, yet the pixel value seem way off :confused: So apparently (if my interpretation happens to be correct) ADC noise is only a minor factor in limiting Dmax, instead the seems to be some form of base fog in every flat bed scanner ...

 

[Doug Fisher]

Thanks, Chris. We all appreciate your posts. I usually learn somthing new from your threads!

Doug

 

[pellicle]

Hi

Now that data set makes me really curious: At the darker sections you claim a standard deviation of 1.4 - 2 %, yet the pixel value

I report what I see in photoshop


click figure for larger view

seem way off :confused: So apparently (if my interpretation happens to be correct) ADC noise is only a minor factor in limiting Dmax, instead the seems to be some form of base fog in every flat bed scanner ...

well, my interpretation is that this 'base fog' is the limits of noise on the conversion of the analog signal from the array of sensors. There is of course some limit at which the sensor can no longer discriminate or detect darkness.

Perhaps the analog gain is too much for the system.

However there is another theory I am intending to persue. This gentleman has explored blacking out the interior of the scanner to reduce extraneous light (flare) from the interior. Grab a nice drink and sit back to read his post here.

Aside from that there is probably some scatter along the light path (including inside the scan head itself).

When I get back to Oz I'll be taking the toolkit to my Epson in view of expanding its black levels.

Importantly (and why I have not yet done it) these levels are of greatest significance to E-6 film users (though they must also contribute to problems in Blue channel noise in C-41), which represents only a tiny portion of what I scan; Black and White in the main. The density levels there seldom go past the 14 step (at least not on my negs)

I can make the raw scan available to you if you wish.

 

[Rudeofus]

Thanks, pellicle for posting these screen shots. To my knowledge these Stouffer wedges you used for your test have 21 strips spaced in density steps of 0.15 and cover D=0 all the way to D=3, which in my books is a 1:1000 range. This means that if the first few strips yield pixel values over 200, the strips with the highest numbers should theoretically yield pixel values of 0 or 1. The standard deviation you see in the histograms indicate that the sensor noise adds up to an RMS of 1-3 LSB, whereas the "base fog" is around 18!

From these results I conclude the following (subject to further verification):

• The scanners you analyzed are not noise limited (in terms of Dmax) but shoulder off at higher densities. Sort of like film does, although the underlying process is completely different.
• With a standard deviation of 1-2 LSB, we get about 7 effective bits per color channel and pixel.
• It would be very interesting to find out whether noise can be reduced by scanning at excessive res (6400 dpi) and downscaling in software afterwards. Going from 6400 dpi down to 2100 dpi combines 9 pixels into one, which reduces the standard deviation of the noise by 3. This would buy us at least 1 more bit of color depth, bumping it to almost 9 bits per color channel and pixel.

 

[pellicle]

Hi

To my knowledge these Stouffer wedges you used for your test have 21 strips spaced in density steps of 0.15 and cover D=0 all the way to D=3, which in my books is a 1:1000 range.

that fits the description of the manufacturer and is what I understood too

adds up to an RMS of 1-3 LSB, whereas the "base fog" is around 18!

yes, as the graph seems to indicate it runs out of discernable readings at about step 16 where little change is indicated

From these results I conclude the following (subject to further verification):

It would be very interesting to find out whether noise can be reduced by scanning at excessive res (6400 dpi) and downscaling in software afterwards. Going from 6400 dpi down to 2100 dpi combines 9 pixels into one, which reduces the standard deviation of the noise by 3. This would buy us at least 1 more bit of color depth, bumping it to almost 9 bits per color channel and pixel.

you may be disappointed to learn that I scanned at 3200dpi on my 4990 and downsampled that ... since the target was not large.

I have some other things to do at the moment, but I can repeat this later if you wish. Meantime my last results are available here. I have some space on my server there so it can remain there until some other space requirement pushes it off the stack.

 

[Rudeofus]

you may be disappointed to learn that I scanned at 3200dpi on my 4990 and downsampled that ... since the target was not large.

Wait ... are you telling me that the histograms you showed were based on already downscaled data? That would change any noise estimate considerably ...

Meantime my last results are available here. I have some space on my server there so it can remain there until some other space requirement pushes it off the stack.

Just curious: these are 4990 scans, or are these from the V700? I'd love to get my hand on such a stouffer wedge or something similar at some point and do some experiments with my V700 myself.

 

[pellicle]

Just curious: these are 4990 scans, or are these from the V700? I'd love to get my hand on such a stouffer wedge or something similar at some point and do some experiments with my V700 myself.

4990 ... I do not have a 700 ... the stouffer wedge is cheap to buy, just get one

 

[Rudeofus]

the stouffer wedge is cheap to buy, just get one

Do they have something similar in Europe, too ?

 

[pellicle]

Hi

Do they have something similar in Europe, too ?

I live in Finland ... postage of such a tiny item was hardly expensive, and Finland is perhaps the most expensive nation in the EU to get anything posted to