Epson V700 vs 4990

[Leonard Evens]

I currently have an Epson 3200 and I'm contemplating upgrading. I do b/w and color negative photography exclusively so D-Max is not really a limiting factor for me. My tests show that I get 28-30 lp/mm effective resolution, which is what a perfect scanner sampling at 1422 to 1524 spi would deliver. Ted Harris has said that the V700, scanning at 3200, which is probably how I would use it, delivers at best just under 2000 spi. He and others also say that used this way it doesn't do as well as the Epson 4990. So it would appear that the 4990 might be the better buy. But I have some further questions which Ted or someone else might be able to answer.

First, it was reported by Roger Clark that he couldn't scan a full 4 x 5 frame with the Epson 4990 at 48 bit color depth and 3200 spi. This seemed to be a problem inherent in the scanner hardware/firmware as confirmed by Ed Hamrick, who makes Vuescan. It worked fine at 2400 ppi or 3200 spi at 24 bit color depth. Second, most reports suggest that the 4990 yields an effective resolution of between 30 and 34 lp/mm, which translates to 1524 to 1727 spi. Roger Clark reported signficantly better results, but otherwise I haven't seen anything suggesting the 4990 could approach 2000 spi. Finally, it does appear that there is considerable variation in what is reported about these scanners. Some of this has to do with technique and how it is measured. I think Ted Harris's estimates are probably the most reliable in this respect. But, there may also be signficant variation in how individual units behave.

Can anyone comment further on any of these points?

[Ed Richards]

The Canon 9950 wil scan a 4x5 at 4800. Like the epson, since the real resolution is less than 2000, this is oversampling. It does, however, reduce the noise and improve the scans. You down sample the file directly in Vuescan or or in Photoshop. While the resolution on the scan I contributed to the scanner collection was not as good as I hoped - in retrospect, I think I had a software problem when I did it - if you look at the noise test, it is quite good, even though it did not get into the shadows nearly as deeply as the best scanners did. With black and white negatives, I think it is has plenty of dmax.

There are some tantalizing references in the Viewcamera scanner article this month to the 700 having a real resolution improvement at 6400 dpi. I would like to hear more about that. I think you could use Vuescan in two passes to handle the files - first pass writes a downsized raw file, the second does the corrections.

[MJSfoto1956]

There is absolutely no possible way the Epson 700/750 can scan "real" 6400 dpi -- total marketing hype I'm afraid. My best estimate is 3200 "real" dpi perhaps a bit less. However, the good news is that is more than enough to make a great scan of a 4x5 negative or non-constrasty chrome. So in spite of the incredible marketing exaggeration of its capabilities, it is still a great buy.

J Michael Sullvan
Editor/Publsher, MAGNAchrom
www.magnachrom.com

[Henry Ambrose]

Leonard,

My 4990 works great with the Epson software up to a point. I ran into the 2400 bug that Epson left us in the standard driver (at this res and higher on 4x5 the scans are very magenta - something software is broken and Epson apparently is not fixing it). This really bothers me but it seems to be the norm now-a-days. So, I bought Silverfast for $100.00 to be able to scan at higher resolutions. My 4990 works great with Silverfast. So one solution is to spend the extra money for Silverfast.

If you use the scanning software to adjust your scans to basic correctness, 16 bit may not matter. If you get -good- 8 bit data tranferred to your computer you're set. If your workflow depends on getting raw data out of the scanner or you intend to do lots of manipulation to the files then 16 bit could be important. How are you used to working now?

From my understanding of the V700 and 750, the improvements come with smaller format films. A 6400 ppi scan from 4X5 will choke your machine. (>2GB !!) Smaller film scans give files that you can handle. I think this makes market sense on Epson's part as there are many more people who will want to get improved 35mm scans than there are LF shooters.

Its an interesting situation we're in now with the Epson's - gathering huge files sizes to overcome crappy optics. I guess crappy optics keep the price down. Fortunately computers are pretty fast now so having to over scan and then post process to get sharp files is not that painfull.

[Leonard Evens]

Henry,

I now use Vuescan under Linux. I use the Gimp to edit my scans, and that program is currrently limited to 8 bits per channel. So in principle I could scan at 8 bits per channel without losing anything. Be that as it may, I get the contrast and color very close in the scan. It seems to me that in doing that part at 16 bits per channel I minimize production of gaps in the histogram. Other steps in the photoeditor are less likely to do that. So I would like to do it that way if possible.

Do I understand from what you say that with SilverFast on the 4990 you can scan the full 4 inch width at 16 bits per channel and at higher than 2400 spi? Roger Clark claimed that the scanner just couldn't do it, and it didn't matter which software he used. Ed Hamrick, in answer to a direct question, said it wasn't something he could fix via sofftware.

I also have plenty of old medium format negatives to scan, so the 6400 spi of the V700 might be useful. But my primary concern is getting significantly better than what my Epson 3200 delivers for 4 x 5. If the both the 4990 and the V700 deliver close to 2000 effective spi, that would constitute such an improvement.

[Henry Ambrose]

I understand wanting 16 bit into your computer but if you get the 8 bit right in the scanning software there is nothing to be gained since (we hope) the scanning software worked in 16 bit. In a sense one very good reason to use any scanning software is that it does something for you instead of you getting raw output from the scanner and then manipulating it. Another way of saying this is if you are collecting an 8 bit file from your scanner that requires - a lot of work - you may want to reconsider your scanning software or how you use it. I'm not saying either way is better.

I'll check later this evening about your scanning question. Which is will it do 4x5 at 16 bit and 2400 ppi or higher - right? I'll try that and post the results.

I found the 4990 to be a noticeable improvement over the 4870 I had previously so I'm pretty sure you'll see a definite improvement over your 3200.

[Henry Ambrose]

Leonard,

I just made a 2400 ppi scan in 48 bit RGB (as Silverfast calls what we've been calling 16 bit) of a 4x5 color negative, full frame. It worked fine, taking about 7 minutes from start to popping up on screen. The file is 548.9 MB

[Leonard Evens]

Henry,

I'm afraid i didn't make it clear enough. I knew that it would work at 2400 spi. The question was whether it would work at resolutions higher than that, e.e.g 3200 or 4800 ppi, with 16 bits per channel.

[Sanders McNew]

I found the 4990 to be a noticeable improvement over the 4870 I had previously so I'm pretty sure you'll see a definite improvement over your 3200.

Really? I went from the 3200 to the 4990 and saw no improvement in image quality whatsoever. The upgrade was a BIG improvement because of the 4990's 8x10 inch scanning area for negatives and transparencies. But my 4990 pulls no more detail, in or out of the shadows, than my 3200 did. Alas.

Sanders McNew
www.mcnew.net/portraits

[Ted Harris]

I can't recall that we ever did a test of the 3200 but we have tested the 4870/4990/V700 (V700 scanning at lower than the 6400 setting) and the improvements all seem to be related to things other than resolution and DMax. The resolution is virtually the same and the DMax, which I recall improves marginally from the 4870 to the 4990 doesn't improve at all on the V700. What Epson is doing is stacking one chip on top of another so that they can claim a continuing higher resolution every time they add another chip. It is an honest theotretical claim but has no real world validity as we all now know.

For those who notice improvements from 3200--->4870--->4990--->V Series I suggest that it could be some combination in a change of work habits, increased skill in scanning and software improvements. Both Epson and Microtek have made some real forward strides with their bundled software in the past two or three years.

In answer to Leonard's questions that started this thread yes, you will get nearly 2000 spi when setting the 4990 or V Series to 3200. The first test scan I ran on the V700 was a 4x5 transparency and I didn't notice any issue scanning the entire frame but, thruthfully, I wasn't looking for that and the image I was scanning was still life with a lot of black around the edges. I did scan using Silverfast SE (which came bundled) at 2400 and 3200 but I can’t recall if it was 8 bit or 16 bit as one of them is not possible with SE as packaged. BTW, while I agree that VueScan is good software and that,if you just want to hit the ‘auto’ button it will usually give you performance as good or better than the others I really don’t think it gives you as much control over your scan as you will get from Silverfast Ai ... bu tyou need to determine if it is worth the huge price differential.

Both Michael and I have noticed that the Epson holders seem to engulf a larger than necessary part of the frame thus eliminating some of the actual detail at the edges of the image. Same is true with the holders for the i900/1800f ... the i800 holders seem to be much better in this regard. I sympathize with the manufacturers here as they have a tough tradeoff to make in terms of giving the user enough of a 'lip' that the film won't fall out and still making sure that no image area is lost ... it is a no-win situation for them when they are designing for the 'average user' out there.

Leonard also raised a valid point about the variability of the units from each to each. My gut would tell me that the QA/QC on all the machines in this price range is somewhat loose but we have used 5 different 1800f's, 3 different i800's and some half dozen different 4870's and 4990's and have not found the performance to be significantly different from one machine to another.

Finally, given the low price of these scanners I a delighted by the quality of their performance rather than dismayed that they are not delivering more.

Hope I answered all the questions.

[Frank Petronio]

What Ted said, they are all a tremendous value for what they are. I had bad luck with quality control on the Microteks, and the progression of Epsons I've had make steady subtle improvements. What is limiting is the crappy Epson software and I do need to try Silverfast. I've been using VueScan and it is too bizzare for color work (it's fun for B&W though!)

I'd save my money and buy Imacon scans for the big stuff. I perceive a subtle improvement between the 3200 and 4990 scans, but I am still working with scans from my old 1600 model that are fine and I don't feel the need to rescan old work because the scanners are so awesomely better.

[Henry Ambrose]

Well, I must have had a "bad" 4870 and a "good" 4990 as I used them side by side on the same film and could definitely see a difference. I never owned a 3200 to compare to these two. I'd still assume improvement with a 4990 over a 3200. Saunders found otherwise in his experience.

I'll claim unit to unit variation has something to do with this along with focus height variatioins. If your Epson flatbed (or any other maker) is not doing as well as you think it should, I suggest you experiment with film height from on the glass and in small increments above that. Somewhere in there you may find a sweeter spot.

Leonard, I'll make another scan and let you know what happens. This time 4x5 at 3200, 48 bit just as you wrote in the original post. That's about a 1GB file I think, pretty darn big to work with unless you're making huge prints and really need it. I've been making really nice 24 inch wide prints from 4x5 scans done at 2400 ppi. Those files are pretty hefty for my older Mac.

[Doug Fisher]

>>I'll claim unit to unit variation has something to do with this along with focus height variatioins.<<

I think you hit on the most likely reason for any variation in sharpness you observed.

Doug
---
www.BetterScanning.com

[Leonard Evens]

Thanks for all the information. If indeed either the Epson V700 or the 4990 can deliver close to 2000 spi, that would be a 33 percent improvement in resolution over what I find I can do with my Epson 3200. Unfortunately, it is impossible to know if the tests were done the same way. I really doubt that I was applying a stricter standard than Ted was, so there is a good chance either scanner would be a significant improvement.

I think I am getting about as much out of the Epson 3200 as possible. If you check

www.jamesphotography.ca/bakeoff2004/scanner_test_results.html

you will see that my contribution was the highest scorer for an Epson 3200.

By the way, I'm well aware of the different factors which can affect one's estimate of performance, including film height. I have done some experiments along those lines with my scanner and I've found the effect is relatively subtle, but it is there.

[Michael Heald]

Hello! Thank you form the link for the scanners. I found the Microtek 5900 on it. The MTF looks good, but it falls way down in chromatic aberration. The optics are problematic; now I know where to expect improvement in when I upgrade. Best regards.

Mike

[Kendrick Pereira]

Question from an ignoramus: Please bear with me.
What do "spi" and "ppi" stand for in scanner-speak? Could someone enlighten me?

[I believe "dpi" stands for "dots per inch"]

[xvvvz]

Maybe others have different definitions, but this is what come to mind for me:
spi = samples per inch
ppi = pixels per inch

You may want to visit Wayne Fulton's www.scantips.com website if you are new to scanning. Great information and tutorials.

Doug
---
www.betterscanning.com

[Leonard Evens]

'spi' is a term introduced, I believe, by Ted harris. It means 'samples per inch'. A flatbed scanner typically collects a horizontal line of samples, stores the results in memory, moves vertically a small amount, collects another horizontal line, etc. The details of how this is done may be more complicated, and there are actually three sets of sensors for each position, providing values for each of the primary colors, R, G, and B. The number of samples taken per inch along each horizontal line is one number and the number of horizontal lines per inch is another number and these are usually given in specifying the scanner resolution, e.g., 3200 x 6400. Usually the smaller number is taken to be 'the resolution'. The measured values are stored in memory as (triplets of) numbers called pixels. Hence the term 'ppi' for 'pixels per inch' is also sometimes used. The term 'dpi' which stands for 'dots per inch' is also commonly used by scanner manufacturers for the same thing, but many people consider it a misnomer. ('Dpi' is also used to describe printer resolution, but in that case, it means something quite different. Relating numbers for digital images from scanners or digital cameras to printer resolution is a bit complicated.)

Digital sampling theory says that if you collect a certain number of samples per uniit length, then the maximal resolution you can achieve in line pairs per unit length is half that. Roughly speaking, you need two pixels for each line pair consisting of a black line and the space next to it. Actual scanners never achieve this theoretcal maximum, but some come closer than others. Usually resolution is measured in 'lp/mm' or 'line pairs per millimeter'. But it could just as well be measured in 'line pairs per inch'. The conversion factor from one to the other is 25.4, which is the exact number of mm in one inch. The Epson 3200 scanner, for example, collects 3200 samples per inch in each horizontal line. That would mean it could at best deliver a resolution of 1600 line pairs per inch, which is the same as about 63 lp/mm. That is the best you could expect to resolve if you took a picture of a resolution test chart with a perfect lens, film capable of infinite resolution, and you scanned it with the Epson 3200. In point of fact, my tests, confirmed by some others, suggest that you can at best do something like 28-30 lp/mm. Translated into inches, that would be be 711 - 762 line pairs per inch. Doubling that, you see that a perfect scanner sampling at 1422 - 1524 samples per inch would give you that. So it would be accurate to say that the Epson 3200 functions like a perfect scanner which samples at aprroximately 1500 spi (or ppi or dpi if you prefer).

The loss of resolution occurs because of faults in the optics and in the scanning hardware. But it is important to remember that the sampling frequency in spi does tell you how many samples will be collected, which will determine the size in pixels of the resulting digital image. For example, my scan region is typically 3.74 x 4.72 inches. So scanning at 3200 dpi yields a digital image of size 3.74 x 3200 by 4.72 x 3200 or about 12,000 x 15,000 pixels. However, the above figures suggest that I won't lose any detail in the image were I to resize that image to about 6,000 x 7,500 pixels. Also, I might do better using a scanner with a lower sampling resolution which had better optics and hence came closer to the theoretical maximum.

[bmarcin]

I have a dumb question. I own a 4990 with SilverFast software. I can see where I can adjust the DPI settings using the Silverfast driver but I don't see any setting options for SPI. From what I understand, I don't believe DPI and SPI are one and the same. With all the talk here about using 2400 SPI, 3200 SPI, could somebody tell me how I would change my SPI settings in Silverfast?

Thanks a lot.

Ben Marcin

[Mike Kovacs]

You certainly did a lot better with your 3200 in the bakeoff than me. Any tricks up your sleave? I've since upgraded to the Doug Fisher holders which are a little better than stock, and let me do my 6x9 MF negs.

I go for a V700 or 4990 just based on digital ICE. I am absolutely fed-up with dust spotting for hours on end with the 3200, compared to my Nikon 35mm scanner...

[Leonard Evens]

Ben,

I'm sorry if I made it too complicated. In effect, 'dpi' for a scanner comes down the same thing as 'spi' or 'ppi' . Any distinctions are too subtle to be concerned about in practice. If you want to call it 'dpi', do so. The reason for making a distinction is that 'dpi' is also used in descrbing printer resolution, where it means something very different.

An interesting question, which others might comment on is just what happens when you ask your scanner to return pixels at less than the specified optical resolution. Since you end up with fewer pixels in each line than there are sensors, there has to be some averaging done. Just where is this done: the scanner firmware/hardware or the scanning software? My conjecture is that the hardware will return half multiples of the optical resolution. So a 4800 spi scanner can return 4800, 2400, 1200, etc pixels per inch, and to return something intermediate, the software writer, e.g. Silverfast, has to do some further programming. I can pretty much see what the software would have to do, but I am not clear what the hardware/firmware combination in the scanner would do. For example, apparently the Epson V700 has a multiple lens system and does different things if you scan at 6400 than if you scan at 3200.

[bmarcin]

Thanks, Leonard. This thread raises an interesting question. I had always been told that I really didn't need to scan any of my medium or large-format chromes at greater than 300 dpi because a printer won't "see" or "process" anything beyond 300 dpi; in other words, I wouldn't achieve any significant detail/resolution beyond 300 dpi that would be worth the larger file sizes. So, I've always scanned my chromes using the DPI setting of 300 using my SilverFast driver. I have some very nice color 30x30" prints made from an Epson 9600 printer (not by me) hanging on the walls. However, on this thread there's a lot of discussion of using very high spi (or dpi) settings: 2400, 3200, etc. Am I being mislead about not going beyond 300 dpi on my scans or am I misunderstanding something completely and describing a totally different situation?

Thanks,

Ben

[Ron Marshall]

You should scan such that at your intended print size you will have 300 dpi (or 360 dpi for Epson printers).

For example: You want to make a 16x20 from a 4x5 negative. That is a four times linear enlargement, so you scan at four times 300 (or 360) that is 1200 dpi.

[Doug Fisher]

Ben -

As mentioned previously, the www.scantips.com website by Wayne Fulton has great tutorials which will help you get a grasp on the difference between scanning ppi/dpi and printing ppi.

Doug
---
www.BetterScanning.com

[Henry Ambrose]

Ben,

Watch your settings in Silverfast. There is a thing they call "quality factor" (I think its called) that has to do with offset printing line screens. Set this to "1" to effectively disable it. (1 x any other settings= any other settings)

I second www.scantips.com as a good resource.

A handy way to keep this straight is to consider your print size and the resolution needed by the printer you are using. A 10 inch wide print from a printer that works best when fed at 360ppi, (spi) requires 3600 pixels to print that 10 inch wide print. (360X10=3600) You'd want to scan so your file contained those 3600 pixels (and a little more so you can crop). If you were using 2 inch wide film you'd need to scan it at 1800 ppi (2x1800=3600).

So when wondering about how big to make the scan, think about how big you might want to print your picture. Then figure backwards from there to arrive at the scanning resolution.

Thats the simplest way to work with scanning I've found.

[Leonard Evens]

Ben,

I agree with what the others have said about the relation between print size and resolution of the image. But let me expand.

The image stored in your computer just has pixels, e.g., 8000 x 10,000. You don't get pixels per inch until you specify a size in inches. For scanning it would be the size of the scanned area. For printing, this would be the size of the print. To go from one to the other involves an enlargement factor.

There isn't a direct relation between pixels stored in your computer and dots of ink deposited by your printer. Each dot is a single color while each pixel consists of three numbers, one for each color R,G, or B. The number of different colors potentially stored in a single pixel is quite large and it is simulated by the printer by a collection of dots of different colors. The number of different colored dots can vary from 4 to seven or more.

But the printer, before figuring out how to distribute dots, converts the pixel array it is sent into its native resolution. You may have better results if you send it a simple multiple of its native resolution, i.e., 360 or 720 for an Epson printer. That is because your photoeditor, e.g. Photoshop, can probably do a better job of converting to the desired resolution.

On the other hand, when you scan, the scanner hardware will collect a certain number of samples per inch. To change to the appropriate resolution to send the printer, you either have to combine pixels or interpolate additional pixels. This can be done by the scanning software, but it is probably better to do it with your photoeditor. So most people recommend scanning at the highest feasible scanning resolution and resizing in the photoeditor.

Generally you want to resize to a smaller number of pixels. Increasing the number of pixels doesn't add any new information to the image because it is done by interpolating new pixels each of which is based on its surrounding pixels.

You should also keep in mind the difference between avoiding pixellation and preserving detail. If you send your printer too few pixels, you will be able to actually see the pixels if you look closely enough. Even if you can't see the pixels, they may still have subtle effects if too large. Usually it is recommended that you send a printer at least 240 ppi, but many people are more comfortable with 300 or even more as a lower limit. This is different from the issue of the printer's native resolution.

But pixellation is not simply related to resolution of detail in the print. 300 ppi, corresponds to about 5 lp/mm and a typical person may be able to resolve between 5 and 10 lp/mm in a print at 10 inches. So if you expect people to view your prints that close, you should not go much below that. But whether or not the image you send to the printer has that detail in the first place is another matter. Suppose the combination of lens, film, and scanner yielded a real resolution of 40 lp/mm (or about 2000 ppi). That would be at the size of the frame, about 95 x 120 mm for 4 x 5 . If you change the size of the final image, you have to divide by the enlargement factor. If you expect critical people to view your prints from 10 inches, you would not be able to enlarge by more than a factor of 4. If you anticipate less critical viewers standing further back, you could enlarge considerably more.

Just how the inherent resolution of the image combines with the pixellation is something I've never seen discussed, but you can't expect to do better than either factor by itself would predict. Most likely, the combined effect is even less than that.

[bmarcin]

Well, thanks very much for the education on dpi, spi and ppi. It seems like I misunderstood the business about scaling to size and, whenever I'd set my dpi in Silverfast to 300 dpi, the overall dpi would adjust for the target size. That answers my basic question here - plus I got a lot of other useful info on this thread. Thank you!

Ben