Increasing neg scan quality on the V700

[jgboothe]

Thought I'd post about something I found out recently about scanning negative film with the V700 and Epson Scan. I haven't seen this mentioned anywhere before, so thought it might be useful for some.

As users of Epson Scan will probably know, it doesn't have an explicit option for multi-sampling. This means you don't have the option to reduce scanner noise or increase its dynamic range. As a result, you get a lot of noise when scanning negatives, which can be particularly bothersome in highlights and overexposed frames.

However, I believe that in positive mode, the scanner does use multi-sampling (possibly in conjunction with varying exposure time per sample). Scan times with equivalent settings are at least double in positive mode compared to negative mode. This would make sense, due to the greater dynamic range needed for scanning transparencies, and I can't think of any other reason why scans would take longer in positive mode.

Further to this, I believe the number of samples is changed according to the positions of the highlight/shadow sliders in the histograms in positive mode, with scans taking longer the further you push the shadow sliders to the left. Again, this would make sense because by doing this, you are 'requesting' that the scanner capture deeper shadows. Whether it's just that more samples are being done, or whether it is using longer exposures for the samples, or both, I don't know, but something different is going on.

This would all be pretty dull, except that I have found that this can result in quite a significant increase in image quality for negatives. There is less noise overall, particularly in denser parts of the film, and as a result, tones and colours are rendered more smoothly and low-contrast detail and texture is clearer. You can apply stronger sharpening, resulting in a sharper-looking image, so I'd even say that effective resolution is increased slightly.

The down side is that set-up times and scan times are far longer (around 3-4 times longer), plus you have to do the negative conversion yourself, with all the associated complexities. However, I find that the time I spend optimising scans post-capture, no matter how I do it, is far greater than the scan time anyway, so for me the increased scan time makes no difference.

To get the benefits, I have found it is necessary to set the input highlight and shadow sliders individually for each image and for each colour channel. I set the output sliders to 0 and 255 to get more linear output and set the h/s input sliders to the left and right ends of the image's tonal range for each channel, leaving significant lea-way to avoid clipping. I then save to 48 bit TIF. It's important to keep the 16 bits per channel, to allow for the transformations necessary during conversion to a 'normal' image.

Here is an example:-

The scan on the left was done as a positive, colour controls on, histogram sliders set as above. The scan on the right was scanned as a negative with 'No color management'. Both were scanned at 6400ppi and saved as 48 bit TIFFs, resized to 3200ppi, then the colour transformations done as a series of adjustment layers in Photoshop. I did my best to equalize the colours. They are both scans of the same 35mm film frame. The film was not moved between scans. The negative has quite a long tonal scale, so the highlight area depicted is fairly dense. It is a very sharp image which contains more detail than the V700 can resolve. The film stock was Fuji Pro160S.

Anyway, see what you think. Maybe some of you already use this method. I suspect the other V-series models will behave similarly. These tests were done using Epson Scan II, but I'm pretty sure the V700 behaves the same with the former version. I'd be interested to hear the experiences of others.

I also did some tests using Vuescan, to see if using multi-sampling in negative mode would produce the same results. What I found was that as the scanning resolution is increased, the number of samples it takes is reduced. When scanning at 6400, it appears that only one sample is taken, no matter how many samples you ask for - the scan time is exactly the same whether you choose 1 sample or 8, as are the results. This was somewhat disappointing, as I was considering purchasing it for this feature. I wonder if this is due to some kind of hardware/firmware restriction. I haven't tested its behaviour in positive mode yet.

 

[Alan Edward Klein]

Can you sum up your results in a simple sentence. I'm getting lost in the weeds.

 

[jgboothe]

Ok. You can get significantly better image quality from colour neg film on an Epson V700 and Epson Scan if you scan as a positive. The scanner uses multi-sampling in positive mode but not in negative mode. You get less noise, smoother tones and a bit more detail. Takes longer, but for me at least, it's worth the effort.

 

[lenshood]

Curious if you've tried this with B&W?

 

[Alan Edward Klein]

Ok. You can get significantly better image quality from colour neg film on an Epson V700 and Epson Scan if you scan as a positive. The scanner uses multi-sampling in positive mode but not in negative mode. You get less noise, smoother tones and a bit more detail. Takes longer, but for me at least, it's worth the effort.

I've read years ago that the reason it's smoother with less noise is because the registration of the two scans can't line up exactly. So effectively, when you combine the two images afterwards, the elements blend together including the grains. You'd probably get the same effect using the smoothing slider in post with a single scan.

 

[jgboothe]

Curious if you've tried this with B&W?

No I haven't tried it with black and white yet - my current scanning project is all colour neg. I see no reason why it wouldn't work the same for black and white though. There is no option for scanning as a B&W positive, so you would be scanning in colour positive mode, same as the method I described, so the scanner would behave the same. I'm fairly sure that you would get the same benefits for B&W. When I get chance I'll give it a try.

I've read years ago that the reason it's smoother with less noise is because the registration of the two scans can't line up exactly. So effectively, when you combine the two images afterwards, the elements blend together including the grains. You'd probably get the same effect using the smoothing slider in post with a single scan.

The method I described is a single scan, as in a single pass of the scan head through the frame. I believe that the sample of each line is done several times. Whether these samples line up perfectly, I don't know. It's also possible that the scanner is lengthening the exposure for each sample rather than doing several samples, but either way, the result is less digital noise and a better capture. I don't think there is any smoothing slider in Epson Scan (just a grain reduction option that I'm aware of?), but anything which is software-based is not the same thing. Using the method described gives a genuinely more accurate capture of the film, which gives a better starting point no matter what you do afterwards in software. I think this is clear from the posted examples.

 

[Alan Edward Klein]

No I haven't tried it with black and white yet - my current scanning project is all colour neg. I see no reason why it wouldn't work the same for black and white though. There is no option for scanning as a B&W positive, so you would be scanning in colour positive mode, same as the method I described, so the scanner would behave the same. I'm fairly sure that you would get the same benefits for B&W. When I get chance I'll give it a try.



The method I described is a single scan, as in a single pass of the scan head through the frame. I believe that the sample of each line is done several times. Whether these samples line up perfectly, I don't know. It's also possible that the scanner is lengthening the exposure for each sample rather than doing several samples, but either way, the result is less digital noise and a better capture. I don't think there is any smoothing slider in Epson Scan (just a grain reduction option that I'm aware of?), but anything which is software-based is not the same thing. Using the method described gives a genuinely more accurate capture of the film, which gives a better starting point no matter what you do afterwards in software. I think this is clear from the posted examples.

If the scanner is sampling more than once, then there has to be more than one scan. I believe you called it multi-scanning. How would you get more than one sample per scan?

On the other hand, it could be that the registration does line up or doesn't matter. After all, when you do ICE, there are two distinct scans and the scanner using infrared determines crud and dust that's on the film with the second scan because it isn't there in the regular non-infrared scan.

 

[jgboothe]

If the scanner is sampling more than once, then there has to be more than one scan. I believe you called it multi-scanning. How would you get more than one sample per scan?

On the other hand, it could be that the registration does line up or doesn't matter. After all, when you do ICE, there are two distinct scans and the scanner using infrared determines crud and dust that's on the film with the second scan because it isn't there in the regular non-infrared scan.

I don't know the exact technicalities of it, but it is possible to get multiple samples with a single pass of the scan head. Each row of pixels is captured more than once as the head passes, and these are then combined. This is how it is done most commonly, I believe. It's also possible to do it as multiple passes of the scan head - you can even do this manually. I would say that the former method has less chance of mis-registration. In Vuescan for the V700, for example, if you set multi-sampling to 2, there will only be one pass of the scan head, but the scan will take roughly twice as long.
In terms of ICE, many scanner/software combinations do this as a single pass also. This is how it works with the V-series scanners and Epson Scan, for example. I presume that for each row of pixels, a normal sample is taken, then one IR sample. The scan takes much longer as a result. However, I'm not an engineer so don't know exactly what's going on under the hood.

 

[Alan Edward Klein]

I don't know the exact technicalities of it, but it is possible to get multiple samples with a single pass of the scan head. Each row of pixels is captured more than once as the head passes, and these are then combined. This is how it is done most commonly, I believe. It's also possible to do it as multiple passes of the scan head - you can even do this manually. I would say that the former method has less chance of mis-registration. In Vuescan for the V700, for example, if you set multi-sampling to 2, there will only be one pass of the scan head, but the scan will take roughly twice as long.
In terms of ICE, many scanner/software combinations do this as a single pass also. This is how it works with the V-series scanners and Epson Scan, for example. I presume that for each row of pixels, a normal sample is taken, then one IR sample. The scan takes much longer as a result. However, I'm not an engineer so don't know exactly what's going on under the hood.

ICE scans on my Epson V600 and V850 twice. It takes twice as long to scan then if you do not use ICE. It doesn't sample each spot twice and then move on. It scans the whole image then does it a second time. Frankly, I don;t see how the image can be sharper with two scans any more than taking two digital shots of the same thing makes it sharper. Of course, HDR is another matter. But that has to do with exposure, not sharpness. Of course, if you're happy with what you get, ignore me.

 

[blockend]

It takes twice as long to scan then if you do not use ICE.

Interesting info. I never use ICE. Next time I'll see how much quicker it is with ICE on.

 

[Alan Edward Klein]

Interesting info. I never use ICE. Next time I'll see how much quicker it is with ICE on.

That's an error on my part. I meant to say it takes twice as long when you use ICE since it scans twice. Once with normal light. And the second time with infrared light.

 

[blockend]

That's an error on my part. I meant to say it takes twice as long when you use ICE since it scans twice. Once with normal light. And the second time with infrared light.

That makes sense, thanks for the correction.

 

[fs999]

On my Plustek OpticFilm 120 using iSRD in SilverFast AI Studio also scans twice, but the infrared scan is faster.

 

[brbo]

Further to this, I believe the number of samples is changed according to the positions of the highlight/shadow sliders in the histograms in positive mode, with scans taking longer the further you push the shadow sliders to the left. Again, this would make sense because by doing this, you are 'requesting' that the scanner capture deeper shadows. Whether it's just that more samples are being done, or whether it is using longer exposures for the samples, or both, I don't know, but something different is going on.

I believe the software adjusts the exposure and this should be pretty easy to establish. In positive mode, make the adjustments that result in longer scan times and compare the file to the "standard" scan in positive mode. If the file is brighter the exposure was longer.

 

[jgboothe]

I believe the software adjusts the exposure and this should be pretty easy to establish. In positive mode, make the adjustments that result in longer scan times and compare the file to the "standard" scan in positive mode. If the file is brighter the exposure was longer.

I'm not sure that a scan being brighter is evidence that more exposure is being used. Surely the brightness can just as easily be the result of post-capture processing, or greater analogue gain being applied? I'm pretty sure that's what happens in the negative scanning mode, as scans always take the same amount of time no matter where you set the sliders. Although it would be nice to know what is going on under the hood, whether it's exposure being varied or multi-sampling, the main thing is that the result is a better, cleaner file, with less noise, than you get from scanning in negative mode.

 

[brbo]

I doubt Epson Scan does multisampling and I haven't come across a scanner/scanner software where "analogue gain" doesn't result in longer scan times. So it's not really like bumping up the ISO on the sensor, but simply extending the exposure time.

The effect of longer exposure time is really easy to test with Vuescan that can give you raw file. Multi-sampling in Vuescan on Epson scanners (not sure about other flatbeds) is, well, strange. The penalty of 16x multi-sampling is only about 20% on my computer when scanning the width of a 35mm negative. More (60%) if I select the entire width of the scan bed. This I take as an indication that multi-sampling is doing something as the wider scan line takes longer to transfer to computer and to average. But if I make the exposure 3x as long the 16x multi-sampling should have more time penalty than it does at base exposure, but the time penalty is still the same (20%). Someone brave enough to contact Ed Hamrick for clarification?

BTW, did you also compare the dark areas (areas with low density on negative) and check that there is no clipping in R channel?

 

[Alan Edward Klein]

Each scanner has a dMax that limits how much light you get through the film. One has to assume that manufacturers designed each scanner to maximize this. After all, the higher the dMax the better the machine and the more they can charge. So they set the amplifiers' gain high enough before overdriving and getting noisy and distorting the signal (like shooting at nominal ISO of 100 in a digital camera rather than 800 when it gets noisier). So if you slow down the scan or increase the amp gain, you're going to introduce noise or other issues that the normal design speed and amplifier settings eliminated or reduced. You can't get something for nothing. So I don't see how a second scan gets you anything you can't get with one normal scan and adjustments in post, which is how I work.

Now the scanner's software, Epsonscan, can apply post-scan settings from full auto to just adjusting things like white and black points. That will change what the results look like from a "raw" flat scan. But those adjustments are done after the sensor captured the signal scan just like Photoshop changes a raw image in post-processing.

 

[brbo]

Each scanner has a dMax that limits how much light you get through the film. One has to assume that manufacturers designed each scanner to maximize this. After all, the higher the dMax the better the machine and the more they can charge.

It's dynamic range (Dmax - Dmin) that matters. Dmax alone is easy, you just need a bright enough light and even the worst possible sensor will see through the densest Velvia slide.

You can't get something for nothing. So I don't see how a second scan gets you anything you can't get with one normal scan and adjustments in post, which is how I work.

Multi-pass scan can prioritise the dense parts of the film (longer exposure) in the first pass and the thin parts (shorter exposure) in the second pass. Then combine the two scans into one image. The perfect registration is a problem at reassembling the image, but with increasing computation power this is no longer a big issue.

Now the scanner's software, Epsonscan, can apply post-scan settings from full auto to just adjusting things like white and black points. That will change what the results look like from a "raw" flat scan. But those adjustments are done after the sensor captured the signal scan just like Photoshop changes a raw image in post-processing.

I don't think the Epson scan is just idling the scanner to make you believe it doing something (longer exposure). I don't know what manufacturer would take pride in longer than needed scan times...

 

[Alan Edward Klein]

How does the scanner handle registration issues when ICE is used?

 

[grat]

The epson uses registration marks on the negative holders.

According to Silverfast, "multi-exposure" is two scans, one at a slower speed, to get more exposure.

 

[jgboothe]

I doubt Epson Scan does multisampling and I haven't come across a scanner/scanner software where "analogue gain" doesn't result in longer scan times. So it's not really like bumping up the ISO on the sensor, but simply extending the exposure time.

The effect of longer exposure time is really easy to test with Vuescan that can give you raw file. Multi-sampling in Vuescan on Epson scanners (not sure about other flatbeds) is, well, strange. The penalty of 16x multi-sampling is only about 20% on my computer when scanning the width of a 35mm negative. More (60%) if I select the entire width of the scan bed. This I take as an indication that multi-sampling is doing something as the wider scan line takes longer to transfer to computer and to average. But if I make the exposure 3x as long the 16x multi-sampling should have more time penalty than it does at base exposure, but the time penalty is still the same (20%). Someone brave enough to contact Ed Hamrick for clarification?

I have done some more testing with Vuescan's multi-sample function on the V700. I recorded the scan time for several scanning resolutions and gradually increased the number of samples requested. Timings only included actual scan time - not calibration or scan head travelling time. Results are posted below.

The results strongly suggest that for any given scanning resolution, there is a maximum number of samples which can be taken - more samples can be taken for lower resolutions. For scans of 35mm width, this is 7x at 800ppi, 4x at 1600ppi, 2x at 3200ppi and just 1x at 6400ppi. It may be that for different scan widths, this will change - I haven't tested for this.
Looking at the scans produced, it was fairly clear that quality was increasing as the scan time increased, but then stayed the same after that, so I'm fairly confident in assuming that if there is no increase in scan time, no extra samples are being done. Scan times increased in a linear fashion as extra samples were taken - ie. 2x samples takes roughly twice as long as 1x. 4x takes roughly four times as long as 1x, up to the 'limit' - after which there was no increase in scan times.
The pattern when scanning in positive mode appears to be exactly the same, albeit with much longer scan times overall.

Scan times are given in seconds.

BTW, did you also compare the dark areas (areas with low density on negative) and check that there is no clipping in R channel?

Yes when scanning in positive mode on Epson Scan using the method I described in the original post, there is absolutely no clipping in any of the channels, highlights or shadows, if you set the sliders correctly. You get access to the full range of tones in colour negative film quite comfortably.

I suspect your are right about Epson Scan not doing multi-sampling (though hard to be sure), but it appears that the extra exposure done in positive mode provides substantial benefits for scanning negatives. If I had to guess, I'd say that the negative scanning mode uses a minimal exposure to increase scan speed, resulting in the image data falling into the lower portion of the sensor's dynamic range, giving quite noisy files - especially in denser areas. The sensor seems capable of dealing with the increased exposure used in positive mode without clipping any of the channels, resulting in much less noisy files and better scans.

 

[Alan Edward Klein]

I have done some more testing with Vuescan's multi-sample function on the V700. I recorded the scan time for several scanning resolutions and gradually increased the number of samples requested. Timings only included actual scan time - not calibration or scan head travelling time. Results are posted below.

The results strongly suggest that for any given scanning resolution, there is a maximum number of samples which can be taken - more samples can be taken for lower resolutions. For scans of 35mm width, this is 7x at 800ppi, 4x at 1600ppi, 2x at 3200ppi and just 1x at 6400ppi. It may be that for different scan widths, this will change - I haven't tested for this.
Looking at the scans produced, it was fairly clear that quality was increasing as the scan time increased, but then stayed the same after that, so I'm fairly confident in assuming that if there is no increase in scan time, no extra samples are being done. Scan times increased in a linear fashion as extra samples were taken - ie. 2x samples takes roughly twice as long as 1x. 4x takes roughly four times as long as 1x, up to the 'limit' - after which there was no increase in scan times.
The pattern when scanning in positive mode appears to be exactly the same, albeit with much longer scan times overall.

Scan times are given in seconds.

View attachment 265882



Yes when scanning in positive mode on Epson Scan using the method I described in the original post, there is absolutely no clipping in any of the channels, highlights or shadows, if you set the sliders correctly. You get access to the full range of tones in colour negative film quite comfortably.

I suspect your are right about Epson Scan not doing multi-sampling (though hard to be sure), but it appears that the extra exposure done in positive mode provides substantial benefits for scanning negatives. If I had to guess, I'd say that the negative scanning mode uses a minimal exposure to increase scan speed, resulting in the image data falling into the lower portion of the sensor's dynamic range, giving quite noisy files - especially in denser areas. The sensor seems capable of dealing with the increased exposure used in positive mode without clipping any of the channels, resulting in much less noisy files and better scans.

Thanks for all that testing. Can you explain the sampling numbers? I don't know what they mean? For example, in the first chart, Number of Samples Set is 2, then for 800ppi you have a 24. What does the 2 mean? What does the 24 mean?

Do you have final results in the form of images that we can compare for the various settings? Or at least two that compare a normal single scan with Epsonscan vs the best you can do multisampling with Vuescan?

 

[jgboothe]

Thanks for all that testing. Can you explain the sampling numbers? I don't know what they mean? For example, in the first chart, Number of Samples Set is 2, then for 800ppi you have a 24. What does the 2 mean? What does the 24 mean?

Do you have final results in the form of images that we can compare for the various settings? Or at least two that compare a normal single scan with Epsonscan vs the best you can do multisampling with Vuescan?

The first column is the number of samples 'requested' in the software. The subsequent columns are the scan times in seconds for different scanning resolutions (actual scanning time - it doesn't include calibration or transport of the head to the scanning position). So in your example, 24 is the number of seconds the scan took when number of samples was set to 2 and resolution was set to 800ppi.

In terms of showing examples, I may get around to doing that, but it takes quite a bit of time to do. Thing is though, anyone interested in getting the best quality possible will want to scan at the higher resolutions, and my tests seem to show that Vuescan only offers limited multi-sampling capability at the higher resolutions (and none at 6400ppi, which is what I would use), whereas scanning in positive mode (at least in Epson Scan - probably in Vuescan as well) gives increased quality at any resolution.

As an aside, I did try stacking multiple scans manually (an alternative way of multi-sampling), and for the most part, it produced good results. Registration of detail can drift a little between scans, but overall, the results were better than just a single scan. I will post examples if I have time.

 

[Alan Edward Klein]

Scanning at 6400 is not going to get a better resolution from my own experiments. I scan 2400 which is pretty much the level that will get you the most. Going higher won't give any more details that uprezing can't do for you later.

Without producing a simple scan result against a multi-scanning result, you really can't assume one is better than the other. Many people make this claim. But I've never seen results that prove their claims.