Noise in Epson 4870 scans

[Robert Jaques]

A week ago I purchased an Epson 4870 to scan my 4x5 Velvia transparencys. My initial impressions of the scans were positive. However the other night I made an inkjet and there appeared clearly visable blotchy noise areas in the print. At first I suspected a problem with the printer but on closer inspection of the scan at 200% magnification I realised the noise was in the scan. The noise was worse in the darker areas as should be expected. The slide I scanned was well exposed and not particularly dark. I tried scanning the slide all sorts of different ways 48bit, 24bit, ICE off, and altering the levels. Turning the ICE off reduced the noise slightly. I have also had the same image scanned on a friends 4870 and the noise is similar. I purchased this scanner after reading glowing user reports on this forum and elsewhere. I thought that the Dmax of 3.8 would be more than adequate, my Canon FS4000 35mm scanner only has a dmax of 3.6 and It produces clean scans most of the time.

My questions are 1 Is there any special scanning technique that can be used with this scanner to minimise the noise. 2 Are there any flatbed scanners on the market that have superior noise performance (in a similar price range) or is the Epson 4870 as good as it gets.

I am beginning to regret this purchasing decision and would appreciate feedback from those who are experienced with scanners (and other 4870 users).

Thank You

[Bruce Watson]

I hate to be the one to tell you, but scanner makers lie like crazy about specifications like Dmax. If your scanner would, in fact, read through a density of 3.8, it would be perfectly adequate for your needs. Sadly, the industry is not reporting Dmax based on any kind of standard test. It's all marketing hype.

One of the reasons that the profession flatbeds like the Creos cost so much is they have better chips, and many of them have active cooling to improve performance. Even at that, it is difficult and expensive to get the level of sensitivity and dynamic range needed to scan trannies from CCD scanners.

Given all that, I think the Epson flat beds are an excellent set of tradeoffs. They do a remarkable job for what they are. But they have limits, and scanning trannies not a task at which they excel, IMHO.

[Andre Noble]

I have scanned a couple of slightly out-of focus 4x5 color negs with the earlier Epson 2450 at 2400 dpi using Silverfast and photoshop software for capture/editing respectively. I carefully sharpened them in Photoshop and printed them out to 11x14 Lightjet with professional results at 300dpi.

I also scanned a very dark 4x5 transparency and did not see artifacts in the 8.5x11 inkjet print, which had some very deep dark tones predominant in it.

At this stage of the flatbed scanning evolution, if you are looking at a full 4800 dpi scan at 200% and seeing some slight artifacts such as solarization in B&W neg scans, or weird color artifacts in color scans, this is to be expected, I guess.

On the otherhand, I can't make out any artifacts of 2400dpi scans of 4x5's at 66% magnification -which will still make a huge professional print on a Lightjet.

If you're viewing at 200% magnification, and after sharpening too much, you're going to get some junk in the view.

The most important thing is that you should make your intial scan on the darker side. Then, you also can't make a radical corrections in Photoshop. All subsequent editing such as "levels", "curves", "saturation", "sharpening", etc. must be subtle or weird stuff shows up fast in the image.

[Paul Butzi]

Multi-sample scanning will dramatically reduce noise. That is, adjusting the scanning software so that it samples each pixel repeatedly instead of just one time.

As a rough approximation, noise will fall as the square root of the number of samples - that is, if you sample each pixel four times, you'll have halved the noise level.

I regularly use 8x and 16x sampling on my Microtek 1800f and it produces dramatically superior results to 1x sampling.

If the Epson driver for the 4870 does not have the ability to do multi-sampling, switching to a second party scanner driver like Silverfast or VueScan will get you there.

The difference is dramatic. Don't give up on your scanner until you've given it a try.

[Guy Tal]

Actually it's worse than Hogarth describes. Some manufacturers rate their scanners' Dmax as a mathematical formula based on bit-depth rather than actual measurement, resulting in such meaningless ratings as 4.8. In reality it means the A/D converter can map colors/tones to a given range of values, but not that the mechanical and optical components can actually discern among these colors/tones in the scanned media.
I have not yet seen a consumer-grade scanner that did not produce visible noise in shadow areas. Note that most "glowing user reports" come from casual home users and are based on limited exposure to other options. Most of them are also very subjective rather than a result of a controlled comparison by skilled operators.
I do think your conclusion is right on the money - for the price, the Epson is a fantastic value and fills a much needed gap. It works well for most needs and many users will never need more than it offers. Still - it falls short of high-end products for situations that demand greater precision.

Guy
Scenic Wild Photography

[John Flavell]

Back in the "old days" of digital photography when we used the first generation photojournalism camera, we had a method for removing "some" noise from the images. I still use this method when noise becomes an issue for any image brought into the digital universe. Even 4x5 high-res scans.

I would give this a try before you start to color correct and before you sharpen.

--In PS, convert the image to a Lab file: image>mode>lab color --Under Window in the menu bar, open channel box: Window>channels, select channels in the pop up box.

--Click on the b channel and your image will go to a pasty black and white. --From the menu bar select Filter>noise>dust and scratches. At this point, look into the area where you found the noise before and see if you can detect the artifacts there. Leave the Threshold selection at zero and slide the radius around to try to smooth out those artifacts.

--Click on the a channel and repeat the above.

--Under Image, convert your image back to whatever you desire. Continue with your normal workflow.

I learned this method with the first-generation cameras that were terrible with noise, but the system does seem to work with high-res scans from slides. It became normal procedure in those days so it's normal today.

Another idea with noise reduction, and I still use this, is the CameraBits Band Aide. Although designed for smoothing out noise in high ISO sports and news photos, it can work on any image where shadow noise is a problem. I believe a trial version is available: http://www.camerabits.com

Hope this is helpful.

John Flavell

[Robert Jaques]

Andr'e- I have been scanning at 1200dpi, I tried a scan a 2400dpi to see if it reduced the noise but it seemed to make no difference. Paul- Silverfast 6 came with the scanner, I don't know if this version supports multiscanning but I shall try it. Ive been using the Epson scanning plugin for photoshop to make my scans. John- Thanks for that technique. I have since made another inkjet print from a different slide that has a fairly dark foreground. Noise is still visable but it is nowhere near as bad as the first print, quite acceptable. I am wondering if my scanner just doesn't like that particular slide. I was planning to have my scans printed on real photo paper at a lab. Do noise artifacts show up less on lab prints LEDS, Lightjets etc?

[Brian Ellis]

Paul - Could you provide the settings in Vuescan for multi-sample scanning?

[Paul Butzi]

Robert-

Yes, silverfast 6.0 does allow multi-sample scanning, at least with my Microtek 1800f. Check the manual, it's not easy to find the feature, it's hidden and you need to click on a cryptic button which changes only slightly as you increase the number of samples. On my version of Version 6, there's a column of buttons running down the left side of the preview window. the top button is an 'i', the next one down is a '?', etc. The 9th one down looks like several sheets of paper. That's the one - click it and a very teeny, tiny number in the lower left hand corner of the button will change from 1 to 2, to 4, to 8, to 16, etc. The number is so damn small I have to put on my reading glasses to read the damn number, it's the worst darn interface I've ever seen. Crank it up to 16, do another scan, and see if the noise gets less.

Brian, I don't use vuescan, so I can't help you out. But I'm pretty sure that it supports multi-sample scanning. Darn sure, in fact.

[Graeme Hird]

Brian,

In Vuescan on the input page, there's an option about 2/3 of the way down called "Number of passes". Set the number to anything between 1 and 16.

It's best if the number you choose is an "integer power of two", (if that is the correct term). It should be 1, 2, 4, 8 or 16 with each of those settings allowing Vuescan to "fill a bit" with information.

Using 1 pass will provide 8 bit depth, 2 will fill 9 bits , 4 will fill 10 bits, 8 fills 11 bits and 16 passes fill 12 bits.

You can never get true 16 bit depth information out of these scanners, but it is an improvement.

(By the way, if you are doing multi-pass scanning, select "48 bits per pixel" in the appropriate selection box on the input page and "48 bit RGB" on the output page under "tiff file type".)

Cheers,