How scanners really work?

[AJSJones]

For typical flatbed scanners (high or low-end), assuming 48 bit acquisition , does the software controlling the acquisition
1) alter the electronics part of the sensor array and its output (i.e. alters the "raw" data) or
2) does the scanner array/A/D converter produce "raw" data that the scanner software then manipulates after the acquisition has happened but before the file is written?

For example, if you have a dark transparency (or a dark area you want to extract shadow detail from) and you set the black and white points, based on a preview scan, to the darkest area you can find and a middle gray respectively, will the quality of the shadow detail/noise level be any better than if you did not set them but adjusted the file in Photoshop afterwards?

Alternatively, if one considers the scanner as taking separate exposures of each scan-line and stacking them to give an x-y file, can you alter the exposure? The reason I ask is because I took a few exposures of such a dark area on a light table with my DSLR (Canon 5D) at different exposures and got better quality from the shadows at long exposures than short ones (no surprise there) and, again no surprise, in the shots where the highlights didn't get blown, the shadows were , well, dark. In the long exposures where the shadow detail was good, it was better than from my scans. In the DSLR world, it is possible to combine a bunch of different exposures to maintain shadow detail and highlights into one file and improve the dynamic range (so-called HDR) and I wondered if any control over the exposure happened in scanners, and if so, perhaps that would work to my advantage.

I would try it myself to see, empirically, but my scanner is packed and ready to move house with me. I do recall, trying that when I got it (an Agfa DuoScan2500) and finding that there was little difference between setting the b and W points in the Agfa Fotolook software and acquiring an "unscaled" raw file and doing it in Photoshop. But I'm self-taught and probably missing something important, so I thought I'd ask somewhere where the experts hang out.

I hope my question makes sense

Thanks

Andy

[Peter De Smidt]

Andy, this is a good question. I was looking at http://www.hutchcolor.com/. They make targets for creating color profiles of scanners. The have a pds of general scanner tips at: http://www.hutchcolor.com/PDF/Scanning_Guide.pdf Basically, they recommend producing a "raw" file, a file with no image adjustment in the scanning software. The idea is to then profile the scanner output using a test target, and all scans would be produced with the same settings, and the scanner's icc profile would be attached. They seem to know what they're talking about (in fact the author's of Real World Color Managment recommended them), but I'd be glad to hear from others much more knowledgable than myself.

[AJSJones]

Thanks Peter. That's a useful pdf and refreshes my memory of how I did my IT8 a long time ago and why I always just did a "raw" scan and attached the scanner profile and did the rest in Photoshop!
A sort of corollary to my question was whether any of the 3rd party software package can control the hardware any differently than the one that came with the scanner, and then, if not, why would some scanner software programs be better than others if you can do post the required processing in PS after the fact?

[Peter De Smidt]

Andy,

Vuescan has a control called "analoque gain", with adjustments for red, green and blue. I think (but I'm not positive) that changing these settings change the exposure times for the three color channels. If so, then this would be a hardware adjustment, as opposed to simply applying a curve to the raw scan data.

[Clay Turtle]

Andy, this is a good question. I was looking at http://www.hutchcolor.com/. They make targets for creating color profiles of scanners. The have a pds of general scanner tips at: http://www.hutchcolor.com/PDF/Scanning_Guide.pdf Basically, they recommend producing a "raw" file, a file with no image adjustment in the scanning software. The idea is to then profile the scanner output using a test target, and all scans would be produced with the same settings, and the scanner's icc profile would be attached. They seem to know what they're talking about (in fact the author's of Real World Color Managment recommended them), but I'd be glad to hear from others much more knowledgable than myself.
Thank you for the tip I saved the pdf file so I may pour over it a little more as time permits!
PS. not for sure about what it is being discussed? A recent Astia slide scan didn't scan produce the low light details present in the Positive . . the responses tended toward issue with D max, etc? Now, I hear your conversation and question arises about scanners in general. My scanner, (HP Scanjet 4890) has manual adjustments for color via a color wheel . . . but in relationship to the Dmax issue, my thoughts ran toward use of manual control via pop-up slider bar that is called gamma value? Did dfferent technologies tend to produce different terms or are you using different aspects?

[jetcode]

Andy,

Vuescan has a control called "analoque gain", with adjustments for red, green and blue. I think (but I'm not positive) that changing these settings change the exposure times for the three color channels. If so, then this would be a hardware adjustment, as opposed to simply applying a curve to the raw scan data.

gain stages doesn't change measurement time because measurement is not accumulative like film, gain stages change the level of signal being measured for that channel in the instant it is measured

[Leonard Evens]

There are several questions of this nature, which I've never got clear in my mind. I think one issue is that there is both firmware in the scanner and the scanning software. The firmware does some preliminary processing and presumably the software may be able to modify what is done by sending appropriate commands to the firmware. But just what it can do is difficult to obtain, and it may differ from one scanner to another.

This arose in connection with some of the Epson scanners. Apparently the width that they can scan at some resolutions is limited by the firmware buffer. So some of these scanners, although in principle they can scan at 4800 ppi, can only do so for limited widths. So if they won't scan a 4 x 5 frame at that resolution. In addition, apparently the only hardware scan resolutions are 4800, 2400, 1200, etc., so if you try to scan say at 3200, you end up scanning at 4800 and scaling down. So you run into the same problem. You have to reduce the scanning frequency to 2400 in order to scan 4 x 5. Apparently this is not something a designer of scanning software can change.

[Peter De Smidt]

Professor Evens is right. Unless the scanner makes multiple passes and stitches the results together, like the Scitex/Creo/Kodak high end scanners, it will not be able to scan large negatives at the same resolution as smaller ones. (Someone on the high end Yahoo scanner group pointed this out.) For example, my scanner has an 8000 element ccd and a maximum optical resolution of 5300 spi. Given those numbers, anything over approximately 1.5" wide will not be able to be scanned at an optical resolution of 5300 spi.

[AJSJones]

I'm OK with the resolution from my scanner - I can't see much detail using a loupe (or digital image) that is not present in the scan. However, it appears that scanner software cannot (or has not been written to) take several readings at the same position and sum (or average) them before moving to the next line. Multiple passes are possible but with problems with the precision of the alignments between passes. When I've moved, I'll investigate the comparisons with digital acquisition and various scanner control options and report anything of interest :D

[Peter De Smidt]

With some scanners, Vuescan allows multiple readings per line, as opposed to averaging a couple of scans, with the concominant alignment issues.

[AJSJones]

Thanks Peter - I'll definitely check into that - I think the T2500 might be one of those
[Edit]Doesn't look like it can do multiple samples per pass - SilverFast may also not be able to get it do multiple scans and average those, but software is out there to try these things on a scanner by scanner basis
Andy