What settings are needed to achieve maximum resolution from an Epson V850?

[williaty]

So, speaking of resolution and the V850, I'm getting some results I'm not sure how to explain. Most of the online reviews estimate the true optical resolution as somewhere between 2000 and 2400 dpi. While I was hunting for best focus, I played around with some settings and turned up results that don't perfectly line up with that. Specifically, I get significantly better real resolution scanning at 6400dpi (software setting) and downsampling to 3200dpi in PS (Bicubic downsampling) than I do scanning directly at 3200dpi (software setting). The test image I've been using has an air conditioner in one window and the cooling fins on the back of it make great test bars. They're are much better resolved when scanning at the higher setting. Film grain is similarly better resolved at the higher setting.

Is this simply a case of having to ask the software for 6400dpi to get the most possible optical resolution even if a proper test chart would show the true optical resolution is down in the 2000-2400dpi range?

[RHITMrB]

It may also be a product of the fact that bicubic resizing (Photoshop's default) also sharpens an image.

[williaty]

I know Bicubic Sharper does, but I didn't think Bicubic did. That being said, the differences don't look like increases in actuance (from sharpening) in that actual separation is improving, not just contrast between them.

[Oren Grad]

I've moved this to a thread of its own, since it's a different topic from the "step up" discussion.

[Peter De Smidt]

Note that a flatbed scanner, unless it stitched ala some Creos/Eversmarts, will have fixed resolution in the direction of the sensor. For instance my scanner has 8000 light sensitive areas in one direction. Those 8000 remain the same when scanning 1 inch and when scanning 8 inches. For my scanner, that means that if I do one pass scanning of an 8x10 negative, that resolution is limited to 1000 (i.e.8000/8) in one direction. This also applies to an Epson. It may achieve 2-2400 dpi over, say, 1", but it'll be lower for bigger scans. The only way with these scanners to increase the resolution for big film is to scan in strips.

[williaty]

Note that a flatbed scanner, unless it stitched ala some Creos/Eversmarts, will have fixed resolution in the direction of the sensor. For instance my scanner has 8000 light sensitive areas in one direction. Those 8000 remain the same when scanning 1 inch and when scanning 8 inches. For my scanner, that means that if I do one pass scanning of an 8x10 negative, that resolution is limited to 1000 (i.e.8000/8) in one direction. This also applies to an Epson. It may achieve 2-2400 dpi over, say, 1", but it'll be lower for bigger scans. The only way with these scanners to increase the resolution for big film is to scan in strips.
Can you provide a reference to back this up? That's contrary to all the other explanations I've ever heard. Everything else says it has a fixed linear resolution along the length of the sensor and your resultant scan has a width of however many pixels your film covers. IOW, if the sensor is a 6400dpi linear, then you'd get 6400px across a 1" wide target, 12800px across a 2" wide target, etc.

[Peter De Smidt]

How would it do that without making more than one pass? If the sensor has 6400 light sensing elements, then that's the maximum amount of points it can read from at one time. The stepping motor/linear drive system might be able to make finer steps in the other direction, but that's it. For a long time, the best sensor used in pro flatbed scanners was made by Kodak. It was a 3-line CCD array (red, green, and blue) with 8000 elements/line. If the entire sensor is being used to give maximum resolution over, say, an inch, i.e. all light sensing elements are being used over that inch, how could the scanner give additional info over the next inch? There'd either have to be another sensor, which there's not, or more than one pass would have to be made.

There are different kinds of flatbed scanners. One has a fixed resolution over the whole bed. This is fairly low, as the sensor reads the whole bed each time. Since the sensor has a given amount of light sensitive elements, this leads to a low obtainable resolution, since if you don't use the whole bed, you wasted sensor sites on blank space. Next up, there are some scanners that can switch lenses, such as the Agfa T2500. It gave one max resolution setting for the whole bed, but it could switch lenses and read a higher resolution strip down the middle of the bed. Getting more complicated there's scanners like my Cezanne, which are xy zoom scanners. This means that the sensor can be moved front to back on the scanner, and that the lens is a zoom lens, which can use all of the sensors elements for any size, with 12" wide or 1 " wide, no matter where the film is placed on the scanning bed. No elements are wasted. Obviously, this is a more complicated way to build a scanner than the earlier mentioned ways, and these scanner were very expensive. Finally, Creo/Scitex/Kodak made some scanners that were xy zoom scanners, but they could also scan more than one strip and combine them. They had a patent on that, and so other scanner manufactures didn't follow suit.

The resolution for the V850 is 4800 x 9600 dpi, according to http://www.epson.com/cgi-bin/Store/jsp/Product.do?sku=B11B224201 . That implies that the sensor has 4800 sensing elements per line. To do 4800 over 8 inches would require a sensor with 38,400 sensing elements-per-line. I have not heard of such a line sensor being produced.

[williaty]

How would it do that without making more than one pass?
If it were a linear array with 6400 photosites per inch over the entire width of the scannable area, it would behave like this. Epson says the max width the high-res camera can scan is 5.9". A linear array 37,760 photosites long (5.9" times 6400dpi) would sample the whole width of the scannable area at once and then be stepped down the scan one row at a time. Interestingly, Epson spec sheet says the maximum resolution for a single-shot scan is 37,760px across the short axis, implying that the sensor physically has that many photosites. However, it also mentions "Alternative 6 Lines Color Epson MatrixCCD", so it may actually have 6 lines of 37,760 photosites each, for a total photosite count of 226,560 photosites. A sensor with 37,760 photosites, or even a 6-line one with 226,560, is tiny by today's standards and wouldn't be any big trick to produce with high yield efficiency. This would produce behavior consistent with everything I've ever read about scanners and in conflict with what you said.

Again, can you site specific authoritative sources that show that scanners work the way you say they do? I would be very surprised to find out it's true and it would certainly change they way I understand them, which is why it's important to me to find out if you're right.



If the sensor has 6400 light sensing elements, then that's the maximum amount of points it can read from at one time.
Why would the sensor have only 6400 photosites?

[RHITMrB]

What makes you assume the sensor is exactly 1" long or scans exactly 1" at a time?

[williaty]

What makes you assume the sensor is exactly 1" long or scans exactly 1" at a time?
Yeah that's exactly my question to him as well. Why does he think the sensor is 1" long? Everything I've ever read about scanners says that the camera samples the full width of the scannable area at the stated resolution in one shot, even if there's a lens involved that "condenses" the image so that the physical sensor occupies a smaller space at a higher pixel pitch. What Peter is suggesting would require the scanner to have a zoom lens that could change how much width it "sees" as you change the scan area in software and that would cost a LOT more than just a simple CCD that was the size it needed to be to do the full width in one shot.

[Peter De Smidt]

First off, I'd be happy to be wrong about this. My understanding about the types of line sensors currently used could be mistaken.

"1 inch", "6400 spi", "8 inch".... were simply examples. I didn't say that any given sensor was 1" long.

I did talk about the various kinds of flatbed scanners, right?

My scanner (Screen Cezanne) does have a hard limit to resolution over a given area due to to the nature of the sensor. It can scan at higher resolutions over a narrower area, as per Screen's documentation, and my own experience using a Edmund Optics' chrome-on-glass high resolution target. It is an xy zoom scanner. See: http://www.screen.co.jp/ga_dtp/en/news/pdf/newsbox/vol7_pdf/newsbox_7_2.pdf

But perhaps Epson has made some huge strides in linear ccd maximum size over the years that I don't know about. The way to confirm one way or the other, other than to have an Epson engineer specify which sensor is used, and it's specifications, or someone tearing apart their scanner to find a part number, would be to scan a resolution target two times. First, scan as if it was a 35mm negative. Second, scan as if it was a larger negative but use the same settings for all other aspects of the scan. If I'm wrong, then the resolution resolved would be the same in both cases. If the resolution resolved is less with the larger area scan, than my suspicion would be confirmed.

Consumer scanner manufacturers are cagey about true hardware specifications, and so it can be challenging to know what's really going on in a scan.

[Jim Andrada]

Why would the sensor have only 6400 photosites?

To save money and make a bigger profit????????????????????

[williaty]

$50 pocket camera has ~10M photosites now, I'm sure in a $1k scanner they can afford 1/20th that many (6-line full width).

[axs810]

But that's what it sells for not how much it costs to make

[Jim Andrada]

From the Epson site

Scanner Type: Flatbed color image scanner

Photoelectric Device: Alternative 6 lines color Epson MatrixCCD®

Optical Resolution:

Epson Dual Lens System
4800 dpi and 6400 dpi
Hardware Resolution:

4800 x 9600 dpi
6400 x 9600 dpi with Micro Step Drive™ technology
Maximum Resolution: 12,800 x 12,800 dpi

Effective Pixels:

40,800 x 56,160 (4800 dpi)
37,760 x 62,336 (6400 dpi)
Color Bit Depth: 48-bits per pixel internal / external1

Grayscale Bit Depth: 16-bits per pixel internal / external1

Optical Density: 4.0 Dmax

High Pass Optics:

Anti-reflective optical coatings
High-reflection mirror
Maximum Scan Area: 8.5" x 11.7"

Light Source: ReadyScan LED technology

Scanning Speed:

High-speed mode: 4800 dpi
Full color: 10.8 msec / line (approx.)
Monochrome: 10.8 msec / line(approx.)

[Ken Lee]

Most of the online reviews estimate the true optical resolution as somewhere between 2000 and 2400 dpi.

See this resolution test (http://www.kennethleegallery.com/images/scanning/Epson750MTFNathanPotter.jpg) of an Epson V750 by Nathan Potter (https://www.flickr.com/photos/argiolus/).

The test shows that once we get past 1500 dpi the contrast drops below 50% - a sort of practical cutoff point for usable resolution.

I routinely scan at 2400 dpi (because the menu gives either 1200 or 2400 and I'm reluctant to type in an intermediate number) knowing that there's no usable resloution beyond that point. I tested the higher settings and found no discernible advantage.

One way to get slightly better data is to scan in 48-bit color, discard the Blue and Red channels and use only the Green channel. See http://www.kennethleegallery.com/html/scanning/scanningGreen.php.

My guess is that the scanning lens and/or sensor is optimized for the frequency of green light.

[Peter De Smidt]

$50 pocket camera has ~10M photosites now, I'm sure in a $1k scanner they can afford 1/20th that many (6-line full width).

Yep, but that's an area cmos sensor, a type of sensor that's had tremendous development over the last decade. We're talking about ccd line sensors, and area with much less development. Epson scanners have been made for over two decades. While they have made their scanners faster, changed the light sources, added different holders, and a few other tweaks, people who've scanned resolution targets haven't seen significant improvements in resolution over that time, excepting, possibly, the new scanners with the second lens. That would seem to indicate no huge advance in the line sensors that they use, at least from a resolution standpoint.

[Jim Andrada]

And the pixel count is just part of it - the performance of the lens(es) and the whole optical path, vibration, alignment, etc etc etc all enter into it. I don't know the size of the sensor, but the overall area is a big determinant of cost, as is, of course, the technology. CMOS is more common than CCD because among other things things there are more sites on a CCD array - the actual photo site as well as a holding site that holds the charge until the entire sensor is read out globally, as opposed to CMOS where each site is read out "on the fly" which is why you get the infamous "Jello-cam/rolling shutter" effect for video. CCD is more expensive because it takes a much larger footprint on the wafer for the same"resolution" compared to CMOS and as Peter pointed out the Epson uses a "line" type sensor which is much less common than an "area" sensor which also raises the cost significantly, to say nothing of the vastly lower volume of Epson scanners vs point and shoot cameras. I'd be surprised if the sensor is not a fairly large component of the total cost (and price)

By the way, re Peter's comment about "zoom", the Epson doesn't "zoom" since it only has two scanning resolutions - one for the 5.9 inch central strip, and a lower one for the full width provided by a wider angle lens, so the experiment of scanning a resolution target at 35mm size vs larger size wouldn't show anything - the equivalent test would be to scan the central stripe vs the full width, and we already know that the full width scan is at lower resolution as a consequence of the design.

[Peter De Smidt]

Good stuff, Jim. Thanks.

[The Joker]

given that the rayleigh limit says

at F5.6 327 cycles/m
at F8 229 cycles/m
at F11 166 cycles/m
at F16 114 cycles/m
at F22 83 cycles/m

then given that a LF neg may frequently be created using f22 or smaller and then also considering that normal film is only able to resolve upto 200 cycles/m but more typically 100 cycles/mm and given that only happens at high contrast ratio so in normal everyday use the max reslution you ar ever likely to get on film is say 80 cycles/m but more typically 40 cycles/m, then it seems that you would only ever require 2000 spi from a scanner since you are never likely to have more than that in the neg to start with. And infact the vast majority of your images will only have 40 cycles/m so a scanner that can do 1000 spi should be adequate most of the time.

So why do people go on and on and on about resolution targets which bear no resemblance to real world photography and why are they worried about whether their scanner scans at 4800 spi or 6400 spi when their negs aren't good enough(capable) to warrant that high level of scanner resolving power anyway? :D:D:D

[Peter De Smidt]

It's true that people often overestimate how much detail they really resolve on film, but a scan doesn't necessarily only deal with scene information. Grain rendition might also be important. On my Cezanne, for instance, scanning at higher spi's give's a much better rendition of film grain than scanning at lower resolutions. I scan 35mm film at 6000 spi, and I do so because I ran a bunch of comparison scans.

[The Joker]

just liking the look of what comes out of a scanner is something completely different than resolution.
All flatbeds and I susepct all scanners create their own grain. Its a misguided impression that what you see from a scanner is "film grain". You scanner will fundamentally alter the film grain and there's nothing you can do about that. The grain you see ain't the same as the grain in the neg.
So all scans are a form of interpolation. I'm not saying that's bad, I'm just saying it happens as part of a scan and if scanning at 6000spi gives you scans you like the look of then fine. But it's scan grain you're seeing not actual film grain.

35mm film shot at say F5.6 has the potential for much higher film resolution but only with the right films and lighting. But a 200 cycles/m film could give you 160 cycles/m so in theory that would require a 4000 spi scan at a minimum. So yes 35mm film can sometimes require those high scan resolutions. But I rekon most of the time they won't Becasue it will be rare that you get as much 80 cycles/m. There just isn't the resolution in the film to extract.

[The Joker]

so to answer the OPs question, shoot at f4 using a film with capability of 400lp/mm and in suitable lighting that gives very high local contrast in the fine detail of your subject.
Then you might approach getting your scanners max resolution and actually be able to see it and NOT be lead into the false opnion that the scanner is not capable just because unwittingly you have not taken into consideration the fact that the resolution didn't exist in your film to extract in the first place.

[Jim Andrada]

Good points. Thanks.

[Peter De Smidt]

By "film grain" I mean the coarse pattern seen in an enlargement, following the very common colloquial use, and not the much smaller grains of silver.

[dbarak]

"My guess is that the scanning lens and/or sensor is optimized for the frequency of green light."

I'm only 6 1/2 years late to the party! Anyway, this would imply - but I could be wrong - that Bayer filtering is involved, since there are more green pixels than red or blue. (Note: I'm not an expert.)

Dave




See this resolution test (http://www.kennethleegallery.com/images/scanning/Epson750MTFNathanPotter.jpg) of an Epson V750 by Nathan Potter (https://www.flickr.com/photos/argiolus/).

The test shows that once we get past 1500 dpi the contrast drops below 50% - a sort of practical cutoff point for usable resolution.

I routinely scan at 2400 dpi (because the menu gives either 1200 or 2400 and I'm reluctant to type in an intermediate number) knowing that there's no usable resloution beyond that point. I tested the higher settings and found no discernible advantage.

One way to get slightly better data is to scan in 48-bit color, discard the Blue and Red channels and use only the Green channel. See http://www.kennethleegallery.com/html/scanning/scanningGreen.php.

My guess is that the scanning lens and/or sensor is optimized for the frequency of green light.