PDA

View Full Version : Scanner test results - 9950 v. 1800



Ed Richards
29-Nov-2005, 12:11
We scanned the same negative in the Microteck 1800 and the Canon 9950, then the Canon file was downsampled to match the resolution of the 1800 file and the contrast was adjusted so they matched. No sharpening. You can download the actual tiff files here and draw your own conclusions:

http://biotech.law.lsu.edu/scanner-test.htm

biotech.law.lsu.edu/scanner-test.htm (http://biotech.law.lsu.edu/scanner-test.htm)

Armin Seeholzer
29-Nov-2005, 13:16
And whicch file is from which?
Would be nice to know!

robc
29-Nov-2005, 13:30
I assume a is the 1800f and it is a tad (but only a tad) sharper. Considering that b has been downsampled, which is throwing away detail and softening, then that puts b on a par with a in my book (that is, if you want to scan at 4800 for the resulting file size, which I do for medium format).

If on the other hand, a is the canon, then, considering the downsampling, it leaves the MT standing in my book.

Mike Chini
29-Nov-2005, 14:22
According to the EXIF data, B is the 1800f. Very disappointing assuming the scans were done properly. I'm in the market for a good flatbed and I keep holding out for some up-to-date info on the 1800F. There are just so many conflicting reports on all of them. What a PITA decision!

Ed Richards
29-Nov-2005, 16:40
> Very disappointing assuming the scans were done properly.

You cannot decide if it is disappointing unless you know how far I was from the newspaper.:-) The double page spread of the newpaper is 7/8 inch wide on the negative. You can just read the regular size type with a 20x loupe on the negative, and it is a very sharp negative. It was shot in bright sun at f11 with a Fuji 250 6.3, on Tmax 100, with a high shutter speed, developed in Xtol at 1:3.

I have added a bigger chunk of the 9950, not downsampled and not adjusted, to the WWW site.

Ed Richards
29-Nov-2005, 16:56
My rough calculation is that a double page spread of the WSJ would need to be about 6 inches wide before you could make out type by eye. Not read it, just make it out. That would be a print about 28x35, with a perfect scan or perfect silver print. So we should be able to make at least 20x25 prints that are indistinguishable from perfect scans. But then, how perfect is a perfect scan? Perhaps we need to send the negative on to one of our drum scanning experts and see what the gold standard is. Of course, that might break our hearts and leave us forever dissatisfied with our scanners.:-)

Mike Chini
29-Nov-2005, 19:12
Ed-

Sorry about that line. Didn't mean to imply anything (I find getting excellent scans myself can be very difficult even with experience).

Anyways, here is a link that shows the 1800f performing really favorably against the Epson 4870:

http://www.gnyman.com/Microtek%20ArtixScan1800f.htm

It is really strange that each review (and corresponding examples) of the 1800f varies so dramatically from one to the next. Perhaps there are significant unit to unit variances or maybe there are very specific workflows for this scanner. Nonetheless, thanks for your input. It is always helpful having actual samples to look at!

Ed Richards
29-Nov-2005, 19:32
> Anyways, here is a link that shows the 1800f performing really favorably against the Epson 4870:

But the 4870 is two generations behind the 9950 - a comparison to it does not tell you anything about a comparison with the 9950. He also tried to multi-sample with the 4870, which will further reduce the resolution because it will drift while you are doing it.

My bottom line, and Paul may want to add his views, is that these scanners give comparable results. The Canon is slightly sharper but the Microtek has slightly lower noise. We are both pretty careful about scanning and have spent some time working on getting good scans. My scanner is on all the time so it stays warm, my room temp is pretty stable, and I have it on a really sturdy table on a thin carpet over concrete. The film also matters. I think Tmax 100 has some real advantages with scanning when you want to resolve fine detail.

The Microtek has other advantages: easier dust control; a much faster scan; and better, but not perfect multi-sampling. (Do not believe that Silverfast hype about perfect alignment with their system - it is good, but not that good.)

So far I think I am the only person on the list who has been working with a 9950, and the assumption has been that the Epson is just as good. I do not know if that is true - Canon builds pretty good optics and I bought the Canon because of my previous experience with their equipment.

Paul Butzi
29-Nov-2005, 19:42
Since it was my Microtek scanner, operated by me, let me just chime in with some details.

The 1800f scan Ed posted was done with Silverfast AI 6, with no multi-sampling, resolution set to 1800 dpi, and the black and white points set to their extremes. Sharpening was turned OFF.
The scan was made as a 48 bit color scan, the red channel was selected in Photoshop and the blue and green channels discarded. The resulting grayscale image was converted to 16 bit grayscale mode, and the crop was then emailed to Ed.

The comparison of the Microtek 1800f to the Epson 4870 is interesting but, in my opinion, not as helpful as you might think. We're not told which software was used for the scans, nor are we told what the settings are. The difference between scans with only a software change can be profound.

In particular, in one of the comparisons, the 4870 not only seems to offer better tonal separation but also displays the 'smearing' of features which often accompany with multi-sampling or else just using grayscale mode on a scanner which does not get good registration between colors. It may be perfectly possible that, if properly operated, the 4870 can do substantially better than what is shown here. It's also possible that the 1800f can do better than is shown here. You just don't know.

I am exceedingly leery of looking at scanner comparisons on the web. There are rafts of variables to getting the best possible scan, and typically most of those variables are uncontrolled in those tests. Something as simple as leaving sharpening on for one scanner and off for another can easily lead you to incorrect conclusions.

To my mind, these uncontrolled variables account for the wild differences in scanner reviews and comparisons, and I think this just highlights the fact that an optimized workflow is essential to getting the best possible scan from ANY scanner.

And, for the record, I'm darn impressed by Ed's 9950 scan. Careful examination of Ed's full scan (he sent it to me on CDR) and my full scans, some of which were done with multiple sampling and other variations, leads me to think that the 9950 edges the 1800f on absolute resolution but loses to the 1800f on noise.

Kirk Gittings
29-Nov-2005, 23:16
"leads me to think that the 9950 edges the 1800f on absolute resolution but loses to the 1800f on noise." This would be my conclusion from the tests too.

Just a couple of things that I don't understand from the texts:

Is this from a 4x5?

Paul, Did you do yours on the glass or in the holder? Emulsion up or down? Why the red channel? All my tests show the green channel as sharper in the 1800f.

I would love to see a sample from these that show how they resolve the grain. None of these samples do. That would tell me more about actual scanner sharpness (and stretching, ghosting etc.) than text.

Are we sure that Vuescan does no sharpening when you say no? Regardless of claims this is not always true with all scanner software. For instance on Imacons when you say no they are still actually selecting out the L channel and sharpening it even when you set it at 0.

Ed as you know I sent back 4 9950's because of a very subtle kind of banding in even middle tone areas like skies. This newspaper subject is not good for testing that. If you have a sample that does, I would love to see it.

Paul Butzi
29-Nov-2005, 23:59
Yes, this is from a 4x5 negative that Ed sent to me.

I used the 4x5 negative holder.

I will have to check tomorrow to see if I put negatives in emulsion up or down. I did some tests early on, marked the carrier, and frankly no longer remember what I do.

I picked the red channel because Ed asked me to make the choice that would give highest resolution, and in my tests, that's the red channel. It's interesting that your tests show the green channel sharpest in your 1800f - if I am remembering correctly, the green channel is *least* sharp in my 1800f.

Hopefully Ed will weigh in with a response on Vuescan and sharpening. I haven't tried that software myself, although after corresponding with Ed, trying it is on my long, long, long list of things to check out. I agree that even when you specify no sharpening the Imacon software sharpens the bejeebers out of things.

Ed Richards
30-Nov-2005, 06:41
The nice thing about Vuescan is that it really does what it claims, and gives you more control of the scanner than any other program. It also got a lot better for the Canon over the past 4 months. Earlier versions of vuescan had some problems, but Ed Hamrick keeps tuning it so you get a new version every couple of weeks with fixes for various scanners.

I ran a series of tests to make sure it was not sharpening, starting with the raw output of the scanner and post processing that myself, then comparing it back to the adjusted output from the scanner. Kirk - what software did you use for those 4 9960s? The banding can be a software artifact as well as a scanner problem. The Canon software is not very good for 4x5. I also limit the processing that the scanner does to limit possible artifacts, i.e., I do not try to optimize the scan, I just try to get the data out.

As for grain - that gets us into the interesting problem of the film you use for scanning. Tmax 100 in Xtol 1:3 is so smooth and the grain is so small that there is no chance you are going to resolve the grain with a consumer scanner. This might be the answer to the sky issue - the smoother the sky, the fewer the artifacts the scanner will generate. The most artifacts occur when the grain is right at the scanner resolution limit, so a little more grain might be worse than a lot more grain.

What Paul and I have focused on is the whole workflow - how do you get the best negative for scanning, not how do you get the best scan of whatever you have been shooting in the past. Paul was an established silver printer and I will defer to his explanation of his views on this. For me, I quit shooting 4x5 15 years ago and only got back to it this spring, so I was starting from scratch. I had been shooting with a DSLR and scanning 35mm film, so I had an idea of what would make a good scan.

I tried some traditional film first - Bregger 200 - and the scans were nasty, so I went to Tmax because the emulsion was designed to have maxium smoothness and limited grain. I did not care about any of the "Tri-x look" sort of stuff, because while I like the look of Tri-x, that is on silver and is irrelevant to scanning. What you need for a scan is a good representation of the dynamic range of the light, but you do not need to worry about whether it fits paper. I have no idea whether my negatives would print on silver. A few of the best were mistakes - I grossly over exposed some film of a very high contrast scene with a white building. I got a very dense negative which I am sure would not print very well (maybe with an alt process), but it scanned fine.

The real issue is what is good enough from a scan. Staring at pixels on a screen is not a good way to evaluate a scan, neither is printing a raw file without processing. While it would be better if the scanner did not create subtle artifacts in the sky (if it does), a little blur will smooth that right out. AA's real skill was in the darkroom, his negatives are not very sharp by modern standards. When I look at the large format images I like, only a few depend on ultimate sharpness for their allure. We also have to accept that digital prints are not silver prints - sharp means something different with ink on paper than with silver and optical printing.

Print size also matters - 20 x 24 seems to me that it might be big enough to do everything you want to do with LF, and consumer level scans might be good enough for that, with a good workflow. If you need bigger prints and sharper images for your vision, a drum scan is in your future, or an 8x10. If money was not an issue, I would have everything drum scanned. But having shot more than 100 sheets of Katrina damage over the past month, that would buy me a new car.:-) Here are couple of quick jpgs for fun: biotech.law.lsu.edu/katrina/ (http://biotech.law.lsu.edu/katrina/) I kept them small, so there are artifacts.

I think we spend way too much time worrying about the input and not enough about our digital processing skills and the output. Look at what this guy does with 35mm black and white:

www.petemyers.com (http://www.petemyers.com)

Kirk Gittings
30-Nov-2005, 08:44
"Kirk - what software did you use for those 4 9960s? The banding can be a software artifact as well as a scanner problem. " I used the Canoscan software. I was definitely not impressed with Vuescan at that time. Maybe it is time to try it again. This was just as the scanner came on the market.

"Tmax 100 in Xtol 1:3 is so smooth and the grain is so small that there is no chance you are going to resolve the grain with a consumer scanner." Not true I have no problem doing this with either the 4990 0r 1800f. Some of the prints in my current show are T-Max 100 in numerous developers.

The thing is I know what a good scan is and how a mediocre scan can make your life more difficult. I figured this out the hard way. I spent the last two years making prints for my current retropective show. I have gotten pretty good at this and even teach it at the university level. As I have written many times. Almost any scanner can give you a descent 11x14 with a good work flow but will not make a 16x20 that holds large format quality. That is the quality barrier. From a large format point of view these scanners barely give an adequate 16x20 where even an Imacon scan at 16x20 will show a real difference.

As for Pete Myers. I can tell nothing about his print quality from a website. I do think it is a little humorous to refer to yourself (as he does) though as "one of the most gifted Master Fine Arts Photographers of our times." A google of his name shows up only his website. He lives in Santa Fe. I have lived in New Mexico all my life and been an active fine arts photographerhere since 1970. I live an hour from Santa Fe and I have never heard of him.

Ed Richards
30-Nov-2005, 09:14
> "Tmax 100 in Xtol 1:3 is so smooth and the grain is so small that there is no chance you are going to resolve the grain with a consumer scanner." Not true I have no problem doing this with either the 4990 0r 1800f. Some of the prints in my current show are T-Max 100 in numerous developers.

Maybe the question here is whether your various developers or your use of them are producing bigger grain, and that is leading to grain aliasing and blasting your scans. You should not be resolving grain in Tmax 100 with the 1800, it should be below the resolution threshold. I cannot speak to the 4990, but I do not resolve it with the 9950. You are an accomplished traditional printer and I assume you are scanning negatives you shot with silver in mind - have you changed your processing and shooting since you started scanning, or are you trying to get good scans of silver printing negatives? As the tri-x look reminds us, grain can increase the illusion of sharpness in silver but it just screws up scanning. (It is also the silver version of digital oversharpening, IMHO.)

As for Vuescan, I would try it again without delay. I am not surprised that you sent the scanners back if you used the canon software. If I had not already been using vuescan, I would have sent mine back, and that was when vuescan was not working so well. That said, I think Silverfast will do as well as vuescan if you output the HDR file and post process it yourself. I do not have as much faith in the internal image processing in Silverfast, but that could because I never spent enough time with it.

Kirk Gittings
30-Nov-2005, 09:38
I will post an example of the banding. I don't know if a down resed file will show it.

Also lok at my posting here http://www.largeformatphotography.info/lfforum/topic/503992.html

Is that not grain? That is FP4. I can show you an example with T-Max.

Kirk Gittings
30-Nov-2005, 10:47
banding in direction of scan path (long dimension) in 9950. This was one of the better ones and that is why I still have it, because I didn't think the client would notice. But this is not acceptable to me so I rejected 4 of these scanners before I gave up and demanded a refund.

http://www.gittingsphoto.com/Articles/HR%20PRES%2003%20crop.jpg

and full image:

http://www.gittingsphoto.com/Articles/HR%20PRES%2003.jpg

Ed Richards
30-Nov-2005, 11:09
Kirk,

You will get no argument from me about the problems with the 9950 with the CANON software. It will not even do a 4800 bit scan at 16 bits on black and white for 4x5, so you cannot even get a basic scan at 4800 to work with.

All of my observations have been for black and white. If you want to do color, get the 1800. To get good black and white you need to scan at 4800 and downsample to 2400. You cannot do that with color, at least on a windows machine, because the file is too big for the memory space. It might work fine on a Mac with a lot of ram, or a Linux box.

The jpg is not working on the link you sent for the grain, but I will trust you that it is grain, or at least grain artifacts. My question is why you have such large grain on Tmax. FP4 I cannot speak to, but I am pretty sure it has larger grain than Tmax 100.

Kirk Gittings
30-Nov-2005, 12:14
Don't you find Vuescan's interface really funky? The only thing comparable is Imacon's which I hate. Imacon should really open up to SF. That would be a great paring.

Ed Richards
30-Nov-2005, 13:17
> Don't you find Vuescan's interface really funky?

You have probably not looked at it in a while. It has been polished up a bit and there are new features. To me, it is just a very straight forward interface for an electronic instrument, but then I worked in research labs back when interfaces might involve punched cards and switches. Anything that is on a screen looks good to me.:-)

Mike Chini
30-Nov-2005, 16:17
You know the professional film companies like Kodak and Fuji could have made using film much more enticing by investing some money into making a really good, affordable pro scanner (up to 4x5). As it stands now, I get results from my new 5D and scanned 35mm (Minolta 5400) that aren't far behind flatbed scans of larger negs. I really don't have much of a reason to stay with lf film (since all of my prints are done digitally now) unless an affordable solution is produced and I don't see R&D picking up for scanners anytime soon.

Ed Richards
30-Nov-2005, 21:27
One more point that I did not think needed mentioning - but another thread proved me wrong. I scan with the emulsion toward the sensor and it makes a difference. The scanner makers should automatically mirror the image and quit telling people to scan with the emulsion away from the sensor. I am assuming that every one who has been testing scanners ignores the manufacturer's recommendations, but if you scanned with the emulsion up, I recommend trying the test again.