LaserSoft Imaging

In the past I have experienced problems when using the Multi-Sampling option in Silverfast where the multiple scans didn't appear to line up and the image appeared to ghosting where the scans were not aligning. I also had this problem when using MEX although not as often. Eventually part of the problem was tracked down to the scanner not being on a solid surface and the slide holder moving during scans. This seemed to get rid of most of the problem but not all of them.

I am now using Silverfast Ai v6.5.1r3 with an Epson v700 Photo scanner and the problem has started again. This time I think there is a new cause and I would be interested to share this theory to see if a) anyone else has this problem and b) does the theory hold up.

I have been scanning a batch of XPan slides and the problem with blur has started to occur in localised patches on the slide. In one scan the bottom left of the image, covering about a third of the frame has become seriously affected but the other areas of the image are pin sharp.

My theory is that the slides are heating up whilst being scanned (I scan at 28"x10" x300dpi) and sampled 8 times. This is causing them to curly slightly along the edges (something I have noticed when removing them from the scanner) and because the Epson slide holder isn't very secure the slide moves gradually through the scanning process.

To support this theory I have only been able to recreate the problem with cold scan's (i.e. the slides have just been placed in the scanner) and the slides are unmounted (XPan and Medium Format). With my scans being quite large and with the multi-sampling on, they get quite warm.

Any thoughts?

I think your theory about a registration problem of some sort is spot on, as the English say. I'm concerned to hear the holders allow movement of the magnitude you are describing. Have you contact Epson about the holders?

Does this happen with 4 samples as well (i.e. does their seem to be a threshold).

Thanks for your thoughts and comments.

I should have explained myself a little more clearly regarding the film holder. It's actually the negative holder I use for scanning XPan slides and it only has the thin edge to hold the slide in place. Once the slide has heated up and is removed from the holder it has a habit of curling along the length, suggesting to me that the top surface is hotter than the bottom surface. I haven't actually seen the film moving in the holder but with a resolution increase of around 1000% it would only require a very slight movement to cause the effect I am experiencing.

Another point that makes me think this might be my problem is that it appears to be mainly a problem with the first slide that is scanned in each batch of film I place in the holder. I suspect by the time the second and third slides are scanned, the film has expanded from the heat.

Regarding the number of samples, 2x doesn't seem to cause a problem but that could be luck. 4x and more and there is a definite problem. By the time I am up to 16x the first slide in he batch is almost always ruined.

At one point I thought it was ICE that might be the problem but I think it's just that with ICE the scans take longer so there is more heat generated.

Thanks for taking time to respond.

Maybe you have already done this test, but what happens if you do not use ICE?

You mention that you are increasing the resolution by 1000%, is that within the "optical" resolution of your scanner? If it is beyond the optical resolution of your scanner there is interpolation taking place which has an impact on the image quality.

Hello!

I have the problem with "intelligent auto-alignment" algorithm too.

I use Epson Perfection 1670. Latest (6.5.1r3) SilverFast AI Studio (demo). It seems, that flatbed scanners do not care about precision of positioning, and it varies from scan to scan. I did some experiments to find it out.

First of all, i needed to be sure that the film is not moving - i put it directly on the glass (my scanner seems to be sharper, when film is put directly on the glass, emulsion side down), and pressed it down with a standart film-holder, so it's flat. On top of it, i placed two match-boxes, so the scanner cover presses the holder, and put some big fat books on top of the scanner, to hold everything still. Solid as a rock :-)

2x Multi-Sampling works from time to time - sometimes good, sometimes - blurs.
4x does good results even rarely...
8x doesn't work good at all - blur every time.
I don't even mention about 16x...

I did some manual scans then. 8 times pressed the "Scan" button, and saved 8 files. The files were a little different - a small offset was present. But only positional offset, not rotational, so it's easy to align everything in Photoshop, if you want.

But I used an old time-proved application to align and average those scans - Photomatix. It aligned them easily and after I did an average there - I was happy! I had an ideal noiseless scan!
And, if Photomatix has a blur too (on some non-contrast scans it happens) - then you can align images pair-by-pair. (1 with 2 = average, 3 with 4 = average, etc... and then align averaged pairs, etc.... - this always works, guaranteed.)

So, the problem is in an auto-alignment algorithm.

Instead of Photomatix, you could use freeware Autopano SIFT, hugin, Panotools or any other panorama software. ( http://user.cs.tu-berlin.de/~nowozin/autopano-sift/ , http://hugin.sourceforge.net/ , http://www.all-in-one.ee/~dersch/ , http://hugin.sourceforge.net/tutorials/scans/en.shtml) - there is a good alignment algorithm there usually.

The other question is, why Silverfast alignment doesn't work? Autopano SIFT is opensource - they could have used working alignment algorithms from there... Or align pair-by-pair, as I wrote before - it's "easier" for software to do that, instead of trying to aling all 16 images in one pass...

So, we'll have to wait for improvement in the next version of SilverFast, or multisampling is useless otherwise (on flatbed epson scanners particularly). And use third-party software for alignment until then..

P.S. Maybe, there is a feature to scan 16 times to different files automatically? I am tired to push the button :-)

P.P.S. Here is a good algorithm description: http://user.cs.tu-berlin.de/~nowozin/au ... tails.html

vicnaum,

I have an Epson flat bed scanner and I am not having this problem with transparencies or reflective images so perhaps I lucked out and got a scanner with better registration calibration, right?

It sounds like you have come up with a good wish-list item for LaserSoft, what do you think?

Thanks for the input folks. To answer your question RAG about the ICE, it doesn't seem to make a difference, although I did notice that ICE Lite makes a much better job of removing dust than ICE Quality.

vicnaum, I am also very interested in your solution as I also use Photomatix and have in the past tried a few mergings with scans at different exposures. Can you give a little more detail of scan settings in order to run them through Photomatix and reduce noise. Do you make several scans all at the same exposure settings or do you adjust between scans.

Something else that I have noticed when running these tests is that the slight offset (which is deffinately present at x2 scas or above) seems to be localised. It's most prominant where dust can be seen against light sky with no detail but less pronounced when there is detail. This could of course be an optical illusion?

Thanks

Hey,

Have either of you tried out Multi-Exposure(ME)? Perhaps it will serve you much better. I did a comparison between some old scans I had done with Multi-Sampling and found the ME scans looked far superior and took less time.

2rnwhalley, RAG: My Epson 1670 doesn't support multi-exposure.

2rnwhalley:
There is nothing special with the settings... I just do several identical scans, and match them with Auto-Align in Photomatix. All of them, or pair-by-pair - it depends on how Photomatix can align images - good, or not. Then I just do "Average" command in there.

All scans I do are in 48bits/TIFF, so I have a clean true-48bit TIFF in the end. Why can't SilverFast itself do the same job - it's a mistery to me... and a question to SilverFast developers.

RAG wrote: It sounds like you have come up with a good wish-list item for LaserSoft, what do you think?

Do I need to post it through the site form, or they will see it here in the forum?

You might want to post it in that section just to be sure.

vicnaum, thanks for the details I will give this a go. It should also answer my concern about the film moving in the slide holder. I will post an update as to how I get on.

RAG, thanks also for your suggestion about MEX. I have tried this in the past and get poor results. There is a problem with areas of high contrast. You can see an example of this in a previous thread linked below.

silverfast-v-650r5-multi-exposure-still-useless-t4698.html

Thanks again for your help.

Tried merging multi scans in Photomatix. Scans align perfectly but my test wasn't very effective at removing noise. I also think there is a bit of film grain coming in to play as well. Will keep trying.

Thanks

Did you use GAIN during your scans to reduce noise? It might help.

No I didn't use GAIN as I find it makes the images slightly soft. My choise of filter is to apply a very small amount of USM during the scan with quite a few tweaks in the advanced settings.

So far when I use Silverfast Multiple Sampling of 4x samples or more I get very good noise reduction but the image isn't as sharp as it could be.

When I scan without multiple sampling the image is very sharp but there is some noise and grain. If I use GAIN the image becomes softer even on low settings. I don't doubt that I could counter this in Photoshop with some work but I would rather not and anyway most of my scans go off to a stock library who understandably don't want a lot of sharpening.

My best scans so far are where I sample once and then apply Neat Image noise reduction. This gives me good noise reduction (not as good as multiple smapling) and a very sharp image.

Sorry to have gone on a bit and thanks for the suggestion.

I might start a new thread for ideas on how to get the very best high res scans. I am becoming a bit of a perfectionist.

Thanks again

OK, strange experience here...

I have tried out GAIN because I wanted to reduce some dark areas of noise without doing multiple scans. I used the expert dialogue to tweak the Theshhold and Intensity to something a little better for my slide film and scan resolution and got quite good results.

I thought that I would then try some multiple samples whilst also using GAIN. It's still early days BUT the alignment of the scans appears much better than previously. Up at 8x smapling the image is starting to get a little soft but I haven't seen the misalignment of images the way I previously did.

If anyone else with an alignment problem reads this can you try out the GAIN filter also.

Thanks