V750 noise comparison

After scanning 35mm with a Minolta DSDIII, I was very surprised by the amount

of noise from the V750. So, for all those interested, here is a comparison

of: 1)EpsonScan, 2) Silverfast ScanMulti, and just for kicks 3) an Imacon 848.

 

There has been no sharpening or other adjustments on any of the scans. All

the scans were made at 2400ppi. To reveal the noise, I ramped up the gamma

(midtone) to 2.5

 

Feel free to download it and take a closer look. One thing I notice is that

the Silverfast scan has less luminance noise, but more chroma noise than the

Imacon. The Epson single pass is really pretty bad in dense areas, even

without trying to show noise it would still be visible.

 

cheers,

Aaron<div></div>

Full frame...<div></div>

If you have Vuescan it's worth a try. Last I heard Ed Hmarick managed to code some hijack of the stepper motor that allowed for single-pass multiscanning on alot of Epson scanners.

What were you scanning? A slide, right?

 

It looks like you've got some bad posterization, but really shouldn't. Are you scanning in 16 bit mode and not doing adjustments with the Epson software (I don't know the software, but it could be applying changes in the driver at only 8 bit precision).

 

Also, can you increase exposure or gain to get the shadows above the noise floor (without destroying the highlights)?

Hi Roger, I scanned them at 48bit in EpsonScan, Silverfast, and Flexcolor. No adjustment except for a single gamma increase of 2.5 to reveal native noise. Of course, the files can easily be adjusted to reduce visible noise, but the purpose of this comparison is to see how much noise is present in the scan.

 

I am in the process of try VueScan and will add an adendum once I've done a few passes with the Multi-sampling. So far, single pass in VueScan is the same as a *single* pass from both EpsonScan and Silverfast, in terms of noise.

 

cheers,

Aaron

Okay, I see from the caption this is a Provia slide.

 

It wasn't clear, but can you do the gamma adjustment in Photoshop and not in the driver for consistency's sake?

 

Also, those evergreen grey/green transitions look terrible on the top 2 scans. What's going on there?

 

I'm looking at your scans at 200% in Photoshop. The Silverfast scan is taming the scanner noise but turning the resolution into mush. Your best bet is probably to skip the multipass multisampling and use a program like Noise Ninja. Or if you need the shadow detail, try a program that lets you blend scans at multipe exposures (doesn't silverfast do this?)

The gamma adjustments were made in PS. Like I said, no driver adjustments were made. This isn't a sharpness or color test, it's a noise test, and the colors are just a result of the extreme curve used to investigate the shadow noise. See the collaborative scanner comparison on www.largeformatphotography.com for more info.

 

I think the Silverfast scan actually holds up pretty well against the 848, even though the 848 is sharper. Nowhere do I find the Silverfast scan losing shadow detail compared to the Epson single pass, either. Of course, I would never print an image with this curve, and on the normal image all colors and color transitions are as they should be.

I'd suggest blending the images with a program that does alignment and blending of multipe images over multisampling. I use Photomatix combine 2 images feature which works well and preserves detail. http://hdrsoft.com/

It may not be reasonable for the size scans you are doing, however, unless you have a *lot* of ram.

Does Vuescan do single-pass multiscanning with your scanner? If so, see if that works better than the Silverfast implementation.

 

Zoom in on your image and you should see that at 8x you are losing image data and sharpness as they get averaged together (it's clear to me). It may not be obvious at 100% but it is apparent at 200% on the file you supplied.

I'm in the process of trying Vuescan. So far, I'm not thrilled about its lack of histogram. It also has a Long Exposure multi-scan function where two scans are made, one very long, the other at normal exposure and their values are selectively combined to bring out shadow detail. It worked very well, but, as mentioned by Hamrick, it did return artifacts around contrast, e.g. the plant stalk on the left. I have not yet tried its true multisampling function. Again, please ignore any tonal loss or "mush" in middle values. The curve I have applied to show noise makes any such comparison of fine midtone to highlight detail null and void. There is clearly detail in the shadows of the Silverfsat scan that is totally obscured in the single pass scan.

Hmmm. I fail to see any noise but I do notice posterisation, which

can be attributed to either (a) narrower bit depth and dynamic range

of Epson scanner (doubtful) or (b) bad input profile.

 

Are you using a custom profile on your Epson? If so, what software

did you use to produce it?

"So far, I'm not thrilled about its lack of histogram"

 

Vuescan does have a histogram and displays raw, image or white/black point data. I think it's on the tab at the far right. I don't find the long exposure pass useful because of artifacts. Photomatix does a much better job.

So it does Roger, thanks. I've used Epson, Minolta, and Silverfast software for a long time, so Vuescan is taking some getting used to. So far, the Vuescan multisampling is pretty good. It's not as strong as Minolta's built in multisampling, but it is way below the noise level of a single pass scan. I might post a revised comparison with them all once I'm a little more used to Vuescan. Really, I just want to show the native noise of this scanner and how it stacks up against a top of the line machine. Everyone is free to find their own way of reducing it, if necessary.

 

Eugene, it's not positerisation. The shadow level doesn't even break 10 on the unadjusted scan, and I'm lifting it to 60 or 70. Take a look at the very last test box on:

 

http://www.largeformatphotography.info/scan-comparison/

 

to see a similar approach.

Yeah, the Vuescan user interface is unique. Not bad once you master it and figure out what everything really does. If you have questions, feel free to email me as I've been using the program heavily for the past few years and am currently helping debug it.

 

I'd be interested in seeing an unmanipulated crop next to the brightened one for comparison (how dark are these shadow areas in reality?)

 

Best,

Roger

As dense as you can possibly get on a P100F slide and still see some form through a loupe on a light table. That's why I chose this slide. There is detail there, it's just really hard to get. In reality, if I were to print this, it wouldn't even need that much shadow detail, just enough to fill the void, so to speak.

Aaron: I am eagerly awaiting your opinion of VueScan's multisampling capabilities.

 

Roger and others: I did a 8x and 16x multisampling on my v700 and it took the same time (or almost the same) as a 1x scan. Is that normal? (35mm slide film - I do see some improvement in noise but very slight - obvious only after I used the Shadow/Highlight tool at 50/50 twice.)

 

Thanks!

Thanks Petre. I'm still getting comfortable with the interface, and myriad settings, but once I get consistent with the program I'll post comparisons. So far, I can't see a difference in noise reduction between 12 and 16x. That said, Vuescan multisampling will probably be my method of choice since Silverfast multi-pass softens fine detail very slightly... and on a flatbed this is not a good thing.

Here's a comparison of different methods of extending dynamic range with my scanner- a limited dmax dedicated film scanner.

 

I'd welcome any feedback or comments.

http://jingai.com/vuescan2/long%20exposure%20comparison.html