Would it be possible to modify an Epson scanner to get higher-res?

[PaulSchneider]

Hi guys,

why is it that the popular Epson V700 series scanners fail to deliver their claimed resolution of 6000+ dpi?

Why is it that flatbed scanners from the 1990s still deliver superior results?

Is it due to cheap optics?

Is the scanning array a cheap component?

If so, wouldn't it be possible to modify, i.e. "pimp", a stock scanner with higher-end components?

For example, if there would be a crowdfunded pimp my Epson 700 kit for say 1000-2000 USD that would lead to true optical 5000 dpi resolution over the entire flatbed ... wouldnt that be exciting?

Rodenstock or Schneider-Kreuznach would surely develop high res optics to spec if there was a substantial enough minimal order? Maybe 100k or 200k USD?

Kind regards

Paul

[GEM]

Hi Paul,

Why not invest in an EverSmart series scanner? They do possess the Rodenstock lens and they are still superior to any of the new flatbeds on the market. It would probably be less expensive than trying to upgrade the Epson V700 (if it is at all possible).

Michael

[BradS]

Just a hunch but, I'd bet that the optic is not the limiting factor. More likely it is the electro-mechanical positioning mechanism.

[Leigh]

The Epson V750 claims 4800 dpi optical, and that only on 4x5 and smaller formats when using the high-resolution lens.
The V750 has two lenses, the V700 only has one lens.

The lateral resolution of a scanner is set by the pitch of the sensor array, which can only be changed by replacing the array with one of finer pitch.

However, on the longitudinal axis the resolution is determined by a complex mechanical assembly, each component of which has manufacturing tolerances.

Achieving a resolution of 4800 dpi on that axis requires moving the carriage in increments of about 0.0002" per step.
Positional accuracy would need to be 1/4th of that or less, which is 0.00005" or 50 millionths of an inch.
Maintaining that accuracy over a span of 12" to 14" requires some very accurate mechanical components and electronic systems.

Given that 4800 dpi is 189 dots per mm(~100lp/mm) , what would be the advantage of increasing the resolving power beyond that?
Achieving 100lp/mm on film requires a very good photographic system (camera + film holder + film + lens + processing).

I don't know of any combination that would yield 200lp/mm to be scanned in the first place.

- Leigh

[Tom Monego]

Your Epson is a $500 scanner, does as well or better than other $500 scanners. Max ppi is about 2000, possibly 3200 on small film. A 4x5 scanned at 2000 ppi will give a 26x33 (approx) print at 300ppi, that is a good sized print, if you want to go larger and stay at 300ppi, not sure an Eversmart can do that, it is more drum scanner territory. Though I'm sure the Eversmart scan would be better.

Yes the 1990s high end flatbeds, like the Eversmart, are better than the Epson but they were $15-20K machines. I photography, when quality becomes the goal you often run into this mass produced vs industrial quality question.

Could you build a better Epson? I'd get a repair manual and go from there.

Tom

[Ken Lee]

According to this article (http://www.flickr.com/photos/argiolus/6154033421/) - written by Nate Potter, one of our esteemed forum members - there is some resolution at the high end of the scale, but at that level, there is very low contrast. After around 2000 spi, it's really nothing to boast about.

So while the company can claim a high number, it's a bit misleading, since the resolution at the high end is... unusable :rolleyes:

For our purposes as LF photographers, the scanner nicely delivers 1600 spi. In round numbers, this allows for an enlargement of 5X: just enough to make a critically sharp 11x14 from a 6x7 negative, or a 20x24 from 4x5 - with no cropping of course.

[cabbiinc]

Why would you spend thousands to repurpose a flatbed scanner when you can buy scanners that are purposed for what you want?

Have you looked into the betterscanning holders? http://www.betterscanning.com/

[Kirk Gittings]

For our purposes as LF photographers, the scanner nicely delivers 1600 spi. In round numbers, this allows for an enlargement of 5X: just enough to make a critically sharp 11x14 from a 6x7 negative, or a 20x24 from 4x5 - with no cropping of course.

Ken, I've owned two and tried everything conceivable to squeeze resolution out of them for years, and I think those numbers are very optimistic. There is no way I would make a 16x20 print for exhibition from a 4x5 without a drum scan.

[GEM]

Your Epson is a $500 scanner, does as well or better than other $500 scanners. Max ppi is about 2000, possibly 3200 on small film. A 4x5 scanned at 2000 ppi will give a 26x33 (approx) print at 300ppi, that is a good sized print, if you want to go larger and stay at 300ppi, not sure an Eversmart can do that, it is more drum scanner territory. Though I'm sure the Eversmart scan would be better.

Yes the 1990s high end flatbeds, like the Eversmart, are better than the Epson but they were $15-20K machines. I photography, when quality becomes the goal you often run into this mass produced vs industrial quality question.

Could you build a better Epson? I'd get a repair manual and go from there.

Tom

The EverSmart series of scanners can absolutely achieve higher resolutions than 2000 ppi. The EverSmart Pro II has a max optical resolution of 3175 over a 12"x17" scan area. The EverSmart Supreme has a true optical resolution of 5600 dpi and will go head to head against a drum scanner.

You are correct about their original price tag. These scanners ranged from $28K to $48K new. However, they are available on the secondary market for less than $5K and $10K respectively.

Michael

[PaulSchneider]

Ok to clear some things up: i own an iqsmart 2 and dont own a v700. But recently i visited a photo store nearby where they had an epson on exhibited attached to a pc. The v700 is much smaller and also scans much faster than the iqsmart. Also it does not require two persons to carry around and is compatible with the newest oses. Ideally i would like to have an epson sized scanner that can do what an iqsmart can do. So thats why im asking.

I understand that theres no more r and d money spent on rifining the epson but i believe that with todays technology it should be possible to create an epson v700 sized scanner that has usb3 scans an8x10 at 5000 dpi in 10 mims and weighs about 10 kg ... Im dreaming here but i feel that there still would be a market for a modern high end scanner that incorporates cost and technology advancements that have been made in the last 10 years ....

I just thought that if its just the optics it should be possible to order custom lenses by crowdfunding them via kickstarter or a similar site. Or if its the microstepping mechanism it wouls surely be possible to swap out compinents with better parts. There still are many people shooting 4x5 and medium format and im pretty sure that a scanner that has higher resolution and is as compact and easy to use as an epson would be a great hit.

I mean comon were writing the year 2012, it shouldnt be too hard to develop such a dream scanner? Schneider kreuznach recently did a custom version of the fine art 770 xxl lens and i believe the minimum order was 20 pieces, so i believe that must have cost 50-60k to convince them do a custom lens .... So it shouldnt be a problem finding enough people who own an epson 700 and want some laser sharp optics?

[BradS]

Again, the optic is not the limiting factor. It is the mechanics and structure necessary to position the optic with sufficient accuracy. Leigh summed it up very well above.

[PaulSchneider]

Leigh,

So in essence that does mean that one would need a far more accurate microstepping device? This probably couldnt be replaced by hand with higher end components?

What i do not understand is how it isnt possible to use the technology in digital scanbacks to scan film? I mean the seitz digital 617 can scan the projected image of a 4x5 lens in seconds and ceeate a 160 megapixel file. Wouldnt it be possible to create a scanner with this technology?

[Leigh]

The problem is not accuracy. They make gage blocks with accuracies of ±2 microinches.
The problem is tolerance over distance in a manufacturing environment, i.e. mass production.

I'm an engineer and a machinist, so I'm pretty familiar with both sides of this problem.

I'm not familiar with seitz, so I can't address its virtues authoritatively.
However, I can say that a fixed sensor array can achieve far higher performance than a scanner with movable parts.

- Leigh

[rdenney]

In addition to the mechanics of moving the linear CCD array, there is also the optics of the CCD itself, plus the glass between the sensor and the film. It is good glass, but not exactly optical grade considering the price. And the sensor sites in that CCD create fuzzy dots that overlap each other significantly, so that one dot bleeds over into the adjacent one or two dots to some extent. All these overlapping fuzzy spots really reduce contrast at the sensor level. The bleed of those sensor sites on their neighbors reduces the ability to record fine detail significantly, and I think that explains the results Nate got.

I think the high-end flatbeds using a tiling approach with a copy lens (a very good copy lens) rather than sweeping a tri-linear CCD with micro lenses. It seems to me this is the reason VueScan has never supported the high-end flatbeds--managing that tiling process is something Ed Hamrick was never prepared to reverse engineer.

Making an Epson do that would be horrendously difficult.

I think there is more promise in our discussion on how to use a DSLR to scan in tiled sections. This is a similar process to a high-end flatbed, but without the full automation and purpose-built lens and sensor combination. But it can outperform the Epson even with modest equipment.

Rick "wondering if there's a way to make a Nikon holder for 4x5:) " Denney

[Brian Ellis]

The reviews of the 700/750 that I saw around the time it was introduced said about 2200 ppi.

[Ken Lee]

The reviews of the 700/750 that I saw around the time it was introduced said about 2200 ppi.

Yes, it can resolve at that level, but as this chart (http://www.flickr.com/photos/argiolus/6154033421/) shows, it's a number of questionable utility, because the contrast at that point is virtually nil.

[sanking]

I think the high-end flatbeds using a tiling approach with a copy lens (a very good copy lens) rather than sweeping a tri-linear CCD with micro lenses. It seems to me this is the reason VueScan has never supported the high-end flatbeds--managing that tiling process is something Ed Hamrick was never prepared to reverse engineer

Rick "wondering if there's a way to make a Nikon holder for 4x5:) " Denney

The very high resolution of Eversmart and IQ scanners is achieved by a lawn mower approach to scanning, and then stitching with proprietary software. These scanners basically scan in strips about two inches wide, then the software stitches the files. This method, in conjunction with very high quality Rodenstock lenses, gives real resolution close to the theoretical maximum. So far as I know the Eversmart and IQ scanners are unique in the use of X-Y scanning and stitching. This allows these scanners to scan any size material at the maximum optical resolution.

Sandy

[Leigh]

Sorry, Ken, but that chart is meaningless.

Let me illustrate, using a hypothetical 5000 dpi sensor.

Given a test target consisting of 5000 lines per inch, of equal width, alternating white and black...

Carefully align the target, accurately focused on the sensor, such that each line is centered on each sensor cell.

The resulting scan will be a nice alternating pattern of white and black dots, subject to any distortion or aberrations that may exist in the optics and the sensor.

Now move the target laterally so that it's aligned exactly with the sensor cell boundary.

The resulting scan will be a nice uniform gray with absolutely no detail.

There's a lot more to evaluating scanner resolution than a simple contrast chart.

- Leigh

[ericpmoss]

Is there any chance that makers of now defunct scanners would "open-source" their software so it could be updated to the latest OSes? I know it wouldn't be of monetary value to them, but would it hurt if they don't even sell scanners any more? Of course, it's a guess if anyone would care to do the port.

[rdenney]

Is there any chance that makers of now defunct scanners would "open-source" their software so it could be updated to the latest OSes? I know it wouldn't be of monetary value to them, but would it hurt if they don't even sell scanners any more? Of course, it's a guess if anyone would care to do the port.

The only scanners not supported by VueScan already, which is cheap, powerful, and kept up to date, are the high-end flatbeds that Sandy describes as using the lawn-mower approach. Ed Hamrick has apparently said that he is unable to reverse-engineer the way in which those scanners assemble the "stripes."

So, you are talking about a small number of very expensive machines. Perhaps one reason they sell for pennies on the dollar is because of their stale software, and their price would increase if there was a good up-to-date software option. Not wanting to maintain an Apple museum has been cited by lots of folks as a deal-breaker.

But I suspect there are other deal-breakers, too, for most folks. The scanners are huge, heavy, and mechanically and optically complex. They require a lot of physical maintenance beyond updated software, and parts are often scarce. I personally would love to have one, but would I invest several thousand in it, given the risks? Not for me. Some will, or have enough production need for it to pay for itself on their immediate project.

The sheer size of these machines is enough to explain a lot of things, however. They are built with the precision required to move the scanner head accurately--remember they work as do all flatbeds by keeping the film still and moving the scanner head.

Film scanners work by moving the film and keeping the scanner head fixed. That makes them mechanically far less complex, because film is light and flat, and moving it 1/4000" at a time is not unreasonable. Both film and high-end flatbeds use a lens to project a portion of the image onto a sensor in succeeding stripes. The high-end flat bed has to move the sensor/lens assembly across the scanned item, and then shift it over to scan another row of those stripes. But both have a high-quality lens that projects the artwork onto the sensor. The scanner head ends up being relatively large because of the intervening lens and the size of the chunk of film it is reading. With a film scanner, the only movement it needs is for focusing, which is a tiny range of motion. As we have already heard, the IQ-Smart uses a Rodenstock lens of very high quality. The Nikon film scanner has the "ED" in the product name (Super Coolscan LS-8000ED or LS-9000ED), denoting a lens with their special low-dispersion glass as they use in their most expensive camera lenses.

Consumer flatbeds like the Epson do not use a separate intermediate lens. Instead, they mold optics into a linear sensor that is the entire width of the scanning bed, and depend on the close proximity of the sensor head and the film to minimize the errors resulting from such a simple lens. That lens serves as a concentrator, I suspect, rather than as a lens that projects an accurate image. The V750 has a "high-resolution" lens that may shift a corrector (or remove one) in front of the strip to alter the focus slightly for the above-glass holders. It seems to me a plausible speculation that the lenses are anamorphic and molded into a strip of one cross-section, since they do not really need to make a round image for each pass of a single sensel, but I have not looked at one up close. I'd be surprised if they were glass rather than plastic.

If one was going to re-engineer a scanner, I think it would be easier to start with a film scanner and construct a holder that would allow a larger piece of film to be held, and then repositioned orthogonally for additional rows. Then, you could stitch the scanned sections in Photoshop. It would require some considerable mechanical reconstruction to do that, but still far less than what would be required to replace the sensor apparatus in an Epson.

But there are better ways if we are going to do the engineering ourselves, though they still use a fixed high-quality lens and sensor, and move the film.

Rick "suggesting that what high-end flatbeds share with film scanners explains why they work so much better than consumer flatbeds" Denney

[Ken Lee]

Sorry, Ken, but that chart is meaningless....

There's a lot more to evaluating scanner resolution than a simple contrast chart.


Could you elaborate a bit ? How do you evaluate scanners ?

Thanks !

[Leigh]

Probably the best target is the fan, with radiating lines.

An alternative would be a clean edge at an angle to the X/Y grid of the scanner.
Look at that edge under high magnification and measure the step increment.

Either target should be chrome on glass.

- Leigh

[PaulSchneider]

Rick,

whouldn't it be possible to take all this:

1) Vuescan scanner software know-how, input from Ed Hamrick, developing stitching algorithmy

2) Digital Scanback-technology based on line-scan ccds (http://www.betterlight.com/how_they_work.html)

3) Some engineering prowess (cnc'ing a housing with microstepping motor in xy axes including parallel moving lightsource from top/bottom)

4) calling up rodenstock and asking them for the flextight/iqsmart rodenstock lens to be remanufactured

and create a new scanner?

[SURF]

What is limiting the resolution: optics or positioning?

Having seen that: http://farm5.staticflickr.com/4147/4989733373_1d5cc658b1_b.jpg made me think that the limiting factor is somewere in CCD-lens. I have no idea how that lens look like in Epson. In Scitex(Creo-Kodak) it is a normal round lens. Scitex scanners also have much better resolution in scanning direction but it is due to smaller numbers of pixels in there CCDs. I do not know if the Epson's CCD pixel number is the limiting factor or not.

http://farm5.staticflickr.com/4147/4989733373_1d5cc658b1_b.jpg

[Ed Kelsey]

I really have to laugh at these ideas....just like the one about building your own drum scanner. Clueless people !

[rdenney]

Rick,

whouldn't it be possible to take all this:

1) Vuescan scanner software know-how, input from Ed Hamrick, developing stitching algorithmy

2) Digital Scanback-technology based on line-scan ccds (http://www.betterlight.com/how_they_work.html)

3) Some engineering prowess (cnc'ing a housing with microstepping motor in xy axes including parallel moving lightsource from top/bottom)

4) calling up rodenstock and asking them for the flextight/iqsmart rodenstock lens to be remanufactured

and create a new scanner?

Go visit the lounge and read through the thread Frank started on building a cheap drum scanner. I think you'll see that what you have described above is being discussed and even prototyped, though not to the extent you are describing.

Of course, your four steps above depart pretty significantly from the notion of modifying an Epson, which lacks the machinery needed to move the scanner head in that fashion and which would therefore have to be redesigned and constructed from scratch. If it was easy to develop that software to support the older high-end flatbeds using the lawn-mower approach, for example, Ed would have already done it. He's had lots of requests to support those older high-end flatbeds. My sense is that the effort required is so different from what he's already done, and the cost so out of proportion to the number of those scanners in existence, that it would make no business sense to do it. But with the right pile of money, it could be done, of course.

Have you considered the cost of your four steps above, and how that might compare to competing alternatives?

Rick "a typical engineer who sees many digits next to that dollar sign" Denney

[BradS]

I really have to laugh at these ideas....just like the one about building your own drum scanner. Clueless people !

let's not leave out the individual who's gonna make film and wants to know what people want in his new film....completely utterly clueless fantasy.

[BradS]

Rick,

whouldn't it be possible to take all this:

1) Vuescan scanner software know-how, input from Ed Hamrick, developing stitching algorithmy

2) Digital Scanback-technology based on line-scan ccds (http://www.betterlight.com/how_they_work.html)

3) Some engineering prowess (cnc'ing a housing with microstepping motor in xy axes including parallel moving lightsource from top/bottom)

4) calling up rodenstock and asking them for the flextight/iqsmart rodenstock lens to be remanufactured

and create a new scanner?


in engineering it is generally about what is possible but rather what is feasible. Yes, of course, it is all possibly but at a cost that I'm sure nobody would be willing to pay. Not to mention that, as Rick has pointed out, this describes an effort way beyond "modifying an Epson V750". It just isn't realistic.

[PaulSchneider]

I know that it sounds unrealistic at first. I'm just interested in different points of view and I believe that there's always a way of doing something that was thought to be impossible at first if enought will and dedication is in place.

Some companies already have many of the elements stated above. Take Seitz, in Switzerland for example. They have a lot of inhouse CNC experience (they actually build the Alpa cameras), they also sell utterly fast line scan cameras (Seitz 617) (160 megapixels in a few seconds!) and could probably easily create a modern scanner.

I'm sure Rodenstock could custom make some lenses for not too much money involved, especially if they can derive it from designs they've done in the past. They recently finished a 770 fine art xxl lens whose minimal custom order threshold was 20 (from what I've heard).

To simplify things, one could make the scan and lens systen non-zoomable so that only one kind of resolution, say 4000 dpi, could be recorded.

If globally 500 people would buy such a system for 2k USD we would have 1 Million USD in revenue, which might be realistic?

[PaulSchneider]

And don't laugh about the film idea either. Fotoimpex bought all the AGFA machines and could market colour emulsions pretty quickly. It's just that right now with kodak dominating the colour market the price point of a new color emulsion would probably have to be much lower to make an impact and to be commercially successful. Kodak my go bust in a few months, their film business will be sold to an investor which will relaunch the old emulsions after a hiatus of production. Maybe only a few months, mayb a year, who knows, but then maybe other people can fill in the void of color emulsions. Also the impossible project just needs some more r&d to finally create the new polaroid. That being said they're globally opening new stores in big cities and successfully selling instant film of dreaful quality. Who would've thought it would still be possible to make money of the polaroid machines without the polaroid patents and emulsions?

[BradS]

...If globally 500 people would buy such a system for 2k USD we would have 1 Million USD in revenue, which might be realistic?

....not much else to say.

[rdenney]

If globally 500 people would buy such a system for 2k USD we would have 1 Million USD in revenue, which might be realistic?

I don't want to sound unduly negative, but I think just the production costs for those 500 units would exceed a million, not counting all the other expenses of developing it, testing it, writing software for it, and the overhead of getting it to those buyers. That's assuming people pay up front on the off-chance that a product will eventually be delivered, when there is every likelihood that the experiment will fail and they'll get nothing.

A CNC machinist might make $100 an hour, not including the time it took to develop the digital data. Something that sells for $2000 should cost no more than $1000 to make (really, more like $500, when you consider those other costs). That's just a few hours for all the various parts needed. And many of those parts will be machined from castings, or you'll spend all that money buying very large billets, most of which will end up as shavings in the bottom of the machine. How many hours of programming would it take to write the software? I'm thinking somewhere in the thousands, which consumes 10-20% of that million bucks. And that programmer, having been paid in advance, is not sharing the risk. I'm seeing hundreds of hours of machine design before anyone even throws the switch on that CNC machine. I suspect the guts of the current high-end flatbeds are already well protected by patents, too, so there would be cost in avoiding someone else's intellectual property.

And let's remember that the Epson scanner that would give its life for the modification would cost half of that $2000, and most of what's inside the case would be thrown in the trash.

The approaches we are playing with in the thread I suggested you read make use of technologies people with some skills can repurpose from the broad market, or make for themselves. Even with that, automating the solution vastly increases the cost of the experiment.

We know this is true. How much does Alpa or Linhof get for a plain camera, which has only a fraction of the complexity being discussed here? How much does Rodenstock get for a production lens?

Responses like mine are often greeted with complaints about how engineers never accomplished anything really innovative because they were afraid to take risks, blah, blah, blah, and always say no to innovation as a result. Despite that such a sentiment is utterly unprovable outside a few anecdotes, it only works if the inventor is investing his own money.

Rick "sorry, Paul, but a million dollars is a lot of money to give, but not a lot of money to spend on a significant research, engineering and production effort" Denney

[Nathan Potter]

There seems to have been some references to a couple of Epson 750 resolution plots I have made in mid 2011. I should add some additional info here for clarification that was omitted on the plots and may cause confusion.

The first (see SURF post, above) was a test to find out the height sensitivity above the platen as a function of resolution. I scanned a sloped glass mask (wedge mask) with repetitive line pair cells over about 2 inches. Thus I scanned through the plane of best focus. Next the file was put in PS and manipulated in levels to yield a very high contrast slice through the resolution pattern to make it easy to see the line pairs under high mag. on the computer screen. This is what was plotted in both the X and Y direction. Notice that this is a plot at near zero contrast so the resolving power plotted is not at all useful for any conclusions about resolution over a full tonal range. This was only done to determine the height of best focus.

Well, with the best focus established I was then interested in determining the relation between resolution and contrast at key resolutions while also preserving the full tonal capabilities of the scanner. Couldn't find any info about this kind of exercise so I just went ahead and did it as best as I could. Some of the details are listed on the plot. What frightened me the most was the possibility of contrast being messed up by aliasing between the spacial frequency of the resolution plate and the spacial frequency due to the scanner sampling. This I think is what Leigh alluded to in his comment above.
However the plot is not meaningless - it is accurate - rather the interpretation of the plot may be meaningless. I have not suggested any interpretation. As Leigh points out that may be quite complicated due to both hardware and software implementation in the Epson. I don't know what kind of anti aliasing is employed by Epson but I think it has to be there since I've not seen any interference effects in all the scans I've done on repetitive structures (brick walls, window screens, etc.).

Also in answer to Leighs concern, I thought I would see different results when measuring other cells in the resolution array but that was not the case within the limit of normal error. So I began to trust the universality of the initial plot a bit more. Also, in general the absolute values of spi capability seem to correspond to what some others have claimed for Epson performance.

If we look at the plot we see at low spacial frequency a change in slope tending to the horizontal. This is where the fuzzy spot discussed by rdenny above begins to fill the space between lines of the resolution pairs. It corresponds to a fuzzy spot (pixel detection area) ranging from say 25 um to 40 um. As we increase the spacial frequency the Epson pixels don't wholly fit within the space between resolution bars and this yields a fairly linear plot from say 18 to 40 lp/mm (1000 to 2000spi). At higher spacial frequencies the the Epson pixel area is so large compared to the resolution frequency that contrast simply washes out. As Ken Lee points out above, at 2200 spi any contrast is questionable; (at 2200 maybe 5%).

After looking at all the scans that were made last summer and making a couple of additional ones lately I have a lot of confidence that the contrast plot is valid to use as a guide for my scanner. Again as Leigh points out manufacturing tolerances can mean variation between Epson units.

I wonder if any members have seen aliasing effects when scanning with an Epson 750? That would be nice to know.

Nate Potter, Austin TX.

[rdenney]

I wonder if any members have seen aliasing effects when scanning with an Epson 750? That would be nice to know.

Nate, I think I have. I compared an Epson V750 scan at 2400 spi with what I saw on the negative with a microscope, and what I interpreted as grain on the scan has a spatial frequency somewhat less than the actual grain. I could not quantify the difference, but it seemed in the range of half the frequency. That seems to me a case of grain aliasing, though its only effect was to make the image seem grainier sooner when enlarged. Even so, no grain was visible in a 16x20 print. Film was 4x5 Ilford FP4.

Rick "whose microscope does have a reticle to aid in counting grains" Denney

[Nathan Potter]

Rick, this possible grain aliasing is something I just don't quite understand. I've heard others suggest grain aliasing effects also. My gut (and observations under metallurgical microscopes) seem to show the smallest as well as the largest grain clumps are randomly distributed thru the emulsion. Of course where this falls apart is when there is very fine spacially repetitive structures of detail equal to the grain sizes. But then aliasing (interference effects) are related to the image structure rather than a function of grain only. Also others have mentioned seeing a grain enhancement (I think Lenny intimated this when he uses certain small apertures on his Aztek). But could this be properly called something else, a different phenomena, rather than aliasing?

I've always thought the aliasing term was restricted to an interference phenomena, where there is a long range spacial effect leading to a variety of Moire type patterns not present in the original.

In fact randomizing any of the hardware and software variables in a scanner coupled with a random grain pattern in the film should eliminate any possibility of aliasing in the scan. The randomness in the scanner could be accomplished in hardware by displacing the individual pixel sensors ever so slightly orthogonally to the scan bar. Or the address to each pixel could be delayed in time randomly. In any such time delay scheme the digital stream would have to be reconstructed using the randomizing key, so to speak, in order to keep the scanned image in sync. and properly reconstructed.

Anyhow I would just be surprised if the effect of grain aliasing some see is really properly denoted as an aliasing phenomena. Or could it be that the silver grains in some films are not completely random but have some sort of spacial regularity which is enhanced through some sort of randomization failure in the scanner?

Nate Potter, Austin TX.

Just curious.

[SURF]

Thank you Nate.

Notice that this is a plot at near zero contrast so the resolving power plotted is not at all useful for any conclusions about resolution over a full tonal range.
Can you post the scan here before PS adjustments?

Remark about grain, resolition, etc.: usually I select the small uniform part of the scan and upply autolevels to it. At 100-200% scale. All the grain starts to be extravisible and convinient to compare.

Regards
Al

[SURF]

Or could it be that the silver grains in some films are not completely random but have some sort of spacial regularity which is enhanced

I think it is more about the size distribution than space. I feel it like certain "ringing" at the highest optical resolution if the scanner lens is still sharp there.

[photobymike]

Ken, I've owned two and tried everything conceivable to squeeze resolution out of them for years, and I think those numbers are very optimistic. There is no way I would make a 16x20 print for exhibition from a 4x5 without a drum scan.

I make 16x20s matted 20x24 on an Epson 4800 from 645 negs and 4x5 negs scanned on a v750 epson... they look great. I have also printed from drum scans, and yes you can see a little difference at extreme magnifications. Also i have scanned on my friends 9000 nikon....On the print i could not tell which scanner i used. I think that digital scanning is all a compromise of quality and money. If you really want the exhibition quality, make a wet print. Quite honestly if you look at my digital prints or my wet prints, you cannot tell what is scanned or what is silver without a loupe.

About the scanner .... its like fitting a hasselblad lens on your mamiya body. Yea!! it would be sharper. I have wrestled with the Epson sharpness issue for a long time. But i keep the Epson kickin because i scan sometimes 100 negs in a day. Thats what i really like about the Epson scanner "fire and forget"

Do i wish i had a drum scanner ... yea! but i dont think it would make a visible difference in my prints? "not so much" So i live with the short comings of my v750 scanner and my Mamiya camera with cheap lenses and my Calumet with a 135mm Carl Zeiss Jena.

[rdenney]

Anyhow I would just be surprised if the effect of grain aliasing some see is really properly denoted as an aliasing phenomena. Or could it be that the silver grains in some films are not completely random but have some sort of spacial regularity which is enhanced through some sort of randomization failure in the scanner?

Any random variable has a mean and a standard deviation if it is normally distributed (as we would expect this random variable to be). It seems to me that if that mean falls near the sensor spacing of the scanner, you could have an interaction even though the random variable is, well, random. In the case I explored, the lens was an old tessar (an Ilex Paragon) used at f/5.6 to achieve selective focus, so it was not resolved to the grain level at all. I was looking across the width of the image of an iron rail with fairly even tone and no repeating texture. The random nature of the grain, and the lack of a decent grid on my microscope, made it onerous to count the grains, but it was clear to me that the scan showed fewer grains across that detail than the microscope did.

I haven't done the math to see how big the variance has to be to overcome aliasing, and at various differences between the mean of the grain spacing and the constant sensor spacing. But it seems to me that if the grain random variable is fairly tightly distributed, and the mean is close to a harmonic of the sensor's spatial frequency, aliasing would still be possible, with the alias pattern showing some relative randomness.

I can't photograph through my microscope, but the next time I have that photo up in PS, I'll take another look at the negative and see if I can put some quantity to it. My microscope is certainly not good enough or powerful enough to quantify the variance, but I might be able to determine a sample mean to compare with the apparent grains in the scan, and see if there is a relationship to the sensor frequency.

Rick "who mathematical statistics textbook from grad school is in a box labeled 'never again'" Denney

[Nathan Potter]

Thank you Nate.

Can you post the scan here before PS adjustments?

Remark about grain, resolition, etc.: usually I select the small uniform part of the scan and upply autolevels to it. At 100-200% scale. All the grain starts to be extravisible and convinient to compare.

Regards
Al

I have to dig deeply into my disk files but should be able to post before PS manipulation. Only constraint is search time, I think.

I like the thought about grain size distribution per unit area. That would be highly variable as a function of exposure dose and even more so as a function of development time. But to end up with some sort of spacial modulation in density - I'm not so sure.

Rick, I like your disposition of your statistics book. My quantum physics text receive a similar fate.

Nate Potter, Austin TX.

[Tyler Boley]

I never quite understood the aliasing nomenclature, wouldn't that imply a byproduct like a moire from a 2 source interference? Effect on grain from drum scanner always seemed to me more related to callier effect. Tim Vitali's paper remains one of the best on the subjet, o matter the terminology. Have I confused things any?
Tyler

[Gene McCluney]

There "used" to be a few mid-price flatbed film-capable scanners that were a step-up from the Epson, but alas with the contracting film market, they went away. With that said, scanning is an "art" and a "skill" and I have seen stunning scans from Epson scanners by those who are very good at scanning. You don't always have to have the highest optical resolution for a big print, with interpolation, it all depends on the subject matter, and your scanning skill and post-scanning file work in Photoshop.

[Nathan Potter]

I never quite understood the aliasing nomenclature, wouldn't that imply a byproduct like a moire from a 2 source interference? Effect on grain from drum scanner always seemed to me more related to callier effect. Tim Vitali's paper remains one of the best on the subjet, o matter the terminology. Have I confused things any?
Tyler

Tyler, I agree in that I always thought aliasing involved some sort of interference phenomena. But maybe the term is used to cite more general cases beyond the simple mathematical case of interference.

BTW, do you have a reference for the Tim Vitali paper handy?

Nate Potter, Austin TX.

[Tyler Boley]

Here you go.... been a long time since I've gone over it so ask me no questions!!!

http://cool.conservation-us.org/coolaic/sg/emg/library/pdf/vitale/2009-10-vitale-filmgrain_resolution_v24.pdf

[federico9001]

I never saw any decent scan from an Epson scanner, even from the best operators (like me)