When scanning photographic prints, the recommendation is to scan at 300 dpi. You will not get more detail if you scan at higher resolutions, just bigger files. Check out Wayne Fulton’s scanning tips (
http://www.scantips.com/).
Regards,
Rafael
Should it always be TIFF files?
Yes. Tif files do not lose detail. If you save as JPG, the file is smaller, but more detail is lost every time the file is saved.
I would scan at 600 dpi, or for sure in 16 bit. That way, if you do any corrections after scanning, you would have some reserve to lose.
Remember, PS works by throwing away data. It cannot create anything new.
Another advantage at 600 dpi is that if you decide to go larger than scanned size, you have an adequete file and do not need to extrapolate.
My advice on resolution would be to do some math, and figure out what resolution you would need. For instance, if the original is 3×5, and you want to go to 8×10, you are going to have to enlarge by 266% (3 to 8). So if you want nice 300 dpi resolution at that size, you should be scanning at 800 ppi now.
Since Bill is scanning 8x10s and 5x7s, I would think he does not intend to enlarge them much, if at all. Scanning at higher than 300 dpi will not gain much image detail, if any. It depends on the printer used to create the originals, but details smaller than 1/300 inch are unlikely.
Certaily, nothing higher than the 600 that Larry suggests makes much sense. Don’s comments are technically correct, but enlarging a 3×5 print to 8×10 is not gonna be pretty.
Bert
Bert,
"but enlarging a 3×5 print to 8×10 is not gonna be pretty"
Why don’t you think enlarging is a good idea? That is exactly what is done when negatives are used to make prints. It is true that the quality won’t be as high as it would if you started out with an 8×10 print, but if it is a good picture you can scan at a higher resolution to get a larger print.
Keep in mind we are talking about scanning at a higher resolution, not up-sampling. It is ridiculous to think that they would add the additional optical magnifications to scanners if they aren’t going to make any difference in the quality of the scan.
Perhaps you were just making a comment on the quality of 3×5 prints but I have scanned wallet prints at high resolutions and made flawless 8×10 prints.
In the referenced "A Few Scanning Tips", Wayne Fulton asserts that color photo prints have a maximum resolution of 200 dpi. To pick up all of 200 dpi, you must scan at 1.5 times 200 dpi (= 300 dpi).
When scanning color prints at scan resolutions above 300 dpi, then, it follows that you are scanning not for detail but for size—you are essentially upsampling via the scan rather than later via Photoshop or some other image editor. If you do not intend to enlarge the original, there is no point in scanning above 300 dpi and you ought not to do so to avoid needlessly large files. If you do enlarge after scanning at 300 dpi, upsampling will give as good a result as having scanned at a higher resolution.
Some sources, without making a distinction between color and B&W, cite that "photograhic prints" max out at 500 dpi. Fulton states that black and white prints may exceed 400 dpi. This suggests scanning good black and white photos at at least 600 dpi.
George,
"upsampling will give as good a result as having scanned at a higher resolution"
No, it won’t. If you believe that you need a new scanner!
PhotoHelp,
I was more or less trying to say what George said…I too am a disciple of Wayne Fulton. You are correct that scanners can scan at very high resolution. My film scanner gives me 2820 dpi, and I always scan at that resolution. Film detail is limited only by grain size, so it makes sense to scan at the highest resolution possible. Scanning printed material, in my experience, is not the same, because you are limited by the resolution of the printer, and probably by the quality of the paper as well.
I’m not saying you shouldn’t scan at higher resolution. It won’t hurt anything. I just wouldn’t expect too much. I am working on a family photo archive. Some of the B&W prints I have scanned go back a hundred years or more. If I scan at high resolution and blow these up, all I see is a blurred and grainy image. Modern prints are better, but there is a real limitation, in my experience, on how much you can accomplish enlarging a scanned image from a print.
Bert
EDIT; Even with the help of Unsharp Mask!
PH
"upsampling will give as good a result as having scanned at a higher resolution"
Let’s keep the statement in context: Upsampling from an image obtained by scanning at a resolution giving all the detail available from the print being scanned. Yes, scanning below that resolution would be inferior. Comes back to what limits the useful scannning resolution—and it is the inherent resolution of the item being scanned. You can’t improve on a sow’s ear. If the print is only 200 ppi, scanning at 4000 ppi isn’t going to give you any detail you don’t get at 300 ppi (1.5x200ppi).
Well, we still have the problem of lost information if we do any major adjustments after scanning.
It doesn’t hurt anything to do 600 or even 720 dpi at the outset, then downsample when you are sure you have the image nailed.
I have scanned and done a major restoration to a 3×5 print, which was 60 years old. It had lost 25% of the detail in the left side of the face for which I had to reconstruct, then make my client an 11×14.
You can bet that I scanned at full scan resolution 16 bit to start. Big file, yep, but as I worked I could check at final size and verify no dropouts. Perhaps it wouldn’t have made any difference, but I would hate to find out as I approached the end of the session.
And, anyway, the 300 dpi final is not only to preserve original information, but to print at a whole number ratio of the printer.
I don’t know why people are so stingy with acquiring information. It’s the least costly investment, Wayne Fulton not withstanding. Wane provides information which I take as the minimum necessary resolutions, not the max. After all, he does title his piece A FEW Scanning Tips………..
A real example:
In part of my business I deal with greyscale aerial photographs. The supplier maintains the photos in both print and negative format. The supplier will also provide scans of the photos, typically from the file copy (meaning its been handled many times) at resolutions up to 4000 ppi(for a portion of the photograph). Since I can not preview or pre-select the photographs for quality, I always request the complete 10"x10" photograph be scanned at 2000 ppi. I found 2000 ppi was the most economical for price while capturing the majority of the information from the print. If I want even greater detail I will request enlargements from the negative. An enlargement of 8x to 10x gives the best results and I can still scan the enlargement at 600 ppi.
There can be a great deal of detail in a good photograph, and more in a negative but many factors can affect the quality. Sometimes I am disappointed by the quality of the scans, but it always goes back to the prints. If I believe there is excessive damage to the prints I request a new print from the negative. And sometimes there is nothing which can be done, because itΒs just a poor photograph.
Steve
"If the print is only 200 ppi, scanning at 4000 ppi isn’t going to give you any detail you don’t get at 300 ppi (1.5x200ppi)."
Exactly. However the print is not 200 PPI. I could see this argument if we were talking about scanning from magazines or inkjet prints. I was just saying that 3×5 prints you get back from say a 35mm camera are higher resolution than this phantom 200 ppi scan resolution you keep bringing up. Pixels are digital the photo is continuous tone. The second you scan it you are converting it to a format with square pixels. Upsampling is going to extrapolate the data from square pixels instead of getting the actual data from the much higher resolution original. Working like that is no better than using JPG’s for master files.
Wayne Fulton’s opinion means very little to me after looking at that web site! Most of what he talks about addresses 1:1 scan vs print resolution without enlarging the print. I wanted to stop reading when he compared scanning to making a copy negative to enlarge a wallet size photo.
"…the photo is continuous tone…"
Continuos tone of the photo paper grain.
That too George! MY 3×5 print had a pebble surface! Nontheless, I scanned fully.
Steve, have them scan the negs. I have a number of 8×10 contact prints and I can tell you that the scan from the print isn’t as good as the scan from the neg. (Confirming the statement about practical resolution of print material). But to be honest, viewing same size prints up to 11×14, I cannot see the difference. At 100%, same file size, the differences are obvious. At 16×20,(breathtaking, even from a scan) you see it if you look carefully.
Perhaps you were just making a comment on the quality of 3×5 prints but I have scanned wallet prints at high resolutions and made 8×10 prints that look fine.
I once restored a photo my mom found that is about 52 years old. It was taken on a Brownie camera, and is a wallet sized print.
It’s old, there’s cracks and scales and fades. However, two of the children in the picture are now deceased but their mother is still alive.
So I scanned it in at something like 1200 dpi on my desktop scanner, then blew it up to 8 x 10 at 300dpi. I lost the fine-line detail, but was able to fix the cracks and creases, smooth the scales, rebuild my mom’s face and print off a very nice looking 8 x 10.
Sometimes it’s the subject that matters, not the photograph. For the lady who lost her children, having a nicely framed 8×10 of her kids when they were 7 years old meant the world to her. She wasn’t looking at contrast levels or photo quality. Although both were good, I just can’t imagine looking at a picture like that and seeing it for it’s artistic value over sentimental value.
I agree if you have any thought of enlarging in the future to up your dpi.
Standard continuous tone photographs are 304.8 dpi so I always scan at 305dpi unless im expecing to blow the image up to poster sizes.
I would say for what your doing, 300 (or 305) dpi is fine. Keep your original working document a PSD and send a tiff version to print.
"…Standard continuous tone photographs are 304.8 dpi so I always scan at 305dpi unless im expecing to blow the image up to poster sizes…"
Whatever resolution your original photo is, you should scan at a resolution of at least 1.5 times that. This dictum comes from mathematical sampling theory, not from the top of someone’s head. Thus, your 304.8 ppi image should be scanned at 450 ppi (roundoff liberties taken).
Whatever resolution your original photo is, you should scan at a resolution of at least 1.5 times that. This dictum comes from mathematical sampling theory, not from the top of someone’s head.
Could you explain why? I’m always on the look out for new information and would greatly appreciate either an explanation or direction to more information.
I am not the original author of the 1.5 sampling ratio statement, but here is my take on it with a simple example:
Pixel 1 sits next to pixel 2.
If the scanner is lined up so scanner-pixel 1 hits print-pixel 1 dead on, you are perfectly lined up and the scan is as good as the original.
If the scanner and photo are off just 1/100 of a pixel (!?!) then scanner pixel 1 gets pixel 1 and a shade of pixel 2. If pixel one is pure white (255,255,255) and pixel 2 is black (0,0,0) then the scanned pixel is something like (200,200,200)(mean, average, etc.) which means you lost the original pure white.
If you scan at 1.5 you have print pixel 1 and pixel 2 being scanned as: pixel 1 (most of print pixel 1), pixel 2 (some of pixel 1 and 2), and pixel 3 (most of print pixel 2) with the values of (255,255,255), (200,200,200) and (0,0,0). This way, you end up with the original info (255,255,255) and (0,0,0) as well as an extra pixel that helps with antialiasing.
This is a very non-scientific, but hopefully useful explaination.
TTFN!
my law is scan at the highest optical resolution your scanner will allow, then you won’t have to rescan later if you need more. my current scanner does 1200dpi optical. hard drive space is cheap. is there a problem w/this?
I just think it is funny that people are making it sound like photo’s have pixels! Or confusing DPI and PPI as usual.
As for seeing grain. If you are scanning at a high enough resolution to see the grain you are scanning at a resolution much higher than 300 PPI.
Think about this… An average ink Jet printer is capable of laying down 1440 dots per inch. If you look at that print you can see the dots with the naked eye. If you look at a photograph you can stare all you want and you won’t see the grain. Even with a 30x microscope it is difficult to see anything more than the suggestion of grain. Yet 300 PPI square pixels are magically able to capture ALL these microscopic details.
300 PPI is plenty to reproduce an image of the exact same size, but if you need to enlarge the image upsampling is definitely no substitute for a good scan!
Dave,
The only problem with overscanning is speed—the time it takes to complete the scan itself and the ability of your system to edit, filter, and otherwise manipulate the product. This can be a real deterrent if you are working with large images (like the 300MB cited) on any system, much less one lacking in resources. The questioner asked what minimum resolution can be used without sacrificing quality, rather than what is good practice in ordinary circumstances to cover all bets.
Cnicknell,
I took an hour to pore through my poorly catalogued files to no avail. I have to say I am frustrated to have misplaced a great article published by a Sandia Labs physicist in a scientific Journal sometime before 2000. Mike Piper’s explanation could be said to be a layman’s gist of it. I don’t recall the journal, author, or year. The title may well have been "How Much Scanning Resolution is Enough?". Despite all this obscurity, the findings seem to be common knowledge—you don’t have to read much to find you are being advised to scan at at least 1.5 times the specimen’s resolution.
The author’s approach was to analyze what happens when you lay out a field of alternating horizontal black and white lines of the same width and scan vertically across them.. The number of line pairs per inch is the inherent resolution. Scanning at that same spatial resolution will give erratic results depending on the phase of the scanner relative to the field. You are pretty much out of the woods at a ratio of 1.5. A ratio of 2 is even better but is gilding the lilly if you are pressed to minimize scanning resolution. The author demonstrates the effects and confirms them via information theory formulas. Sorry for being a poor librarian.
Or confusing DPI and PPI as usual.
ph, you know i’m not doing that. I actually struggled w/typing that, but didn’t want to drag that issue up, and as that’s the refrence on the scanner, that’s what I used (even though we both know it’s not accurate). Anywho… any comment at my practice of "always scan at the highest optical res. available" (1200PPI)?
my reasoning is if i’m scanning now, grab all the info i can. as i said, hard drive space is cheap. very cheap. I don’t have an inkjet, i have a b&w laserjet (hp laserjet ii!!! :)) and an olympus dye sub (continuous tone). but even more to the point, i use most of my images digitally, so i want as much detail as possible. Also if I ever want to take one to the lab to have blown up again, I want to have all the information possible. Most of my prints are on glossy paper anyway, so there’s not much visible grain.
comments? is this wrong?
The questioner asked what minimum resolution can be used without sacrificing quality, rather than what is good practice in ordinary circumstances to cover all bets.
good point george. sorry to pull y’all away from the discussion. But as for the speed issue, i’d still scan high, save the high res scan, then downsample to the needed resolution.
Dave,
I wasn’t referring to you.
"…An average ink Jet printer is capable of laying down 1440 dots per inch. If you look at that print you can see the dots with the naked eye…"
I’m making an appointment with my opthamologist first thing in the morning!
George, Shannons Sampling Theorm says the sampling frequency needs to be 2x the sampled frequency to avoid aliasing. Sampling Oscilloscopes always used 2x (at least when I was doing samplers) So, the -3db frequency of the instrument was 0.5x the sampling rate. Samplers were always only half as fast as the fastest pulse we could generate. Of course, who knew exactly what the pulse generator was really doing because there was no instrument to look at the sampling pulse. Soooo..
Those were the days, my friend………..
any comment at my practice of "always scan at the highest optical res. available
Sounds like a good idea. Unless your scanner goes up to 11.
Mine does 3200 d.p.i. (that’s what they call ’em on the box), and if I want fewer pixels, I throw them away later. Pixels are recyclable.
and if I want fewer pixels, I throw them away later. Pixels are recyclable.
that’s my feeling. thanks.
A photo print’s maximum "resolution" is limited by the paper-ink-printer system’s ability to lay down a grid of alternating equally wide black and white lines recognizable by any optical device (magnification permitted). For whatever reason, that limit is allegedly about 500 line-pairs per inch. And, obviously,no scanner is going to milk more resolution out of any print so limited. It is also obvious that this resolution will be degraded when three different color inks are involved rather than just black, so a color resolution limit of 200 or 300 lpi is not an unreasonable expectation given a black-white limit of 500. The bone of contention, therefore, is this 500lpi limit for photographic prints. PH challenges it. I can’t refute the challenge, but I can say the limit is consistent with my observations, experience, and the general literature.
Bill Brammer no longer cares.
But I’m finding this interesting, so, carry on, fellas.
well what do you do phos? scan at the needed res, or scan at the highest possible, anticipating possible future need?
Depends on the job and the client.
If it’s a one-off photo restore, I’ll scan for final output size. If I have to do some REALLY fine work, I might bump the resolution a bit, but generally, 300 ppi at final size is good. No sense working on a 16-bit/channel image at super high resolution for a brochure if it’s just gonna be reproduced on a CLC-1000 color laser printer.
If it’s for a regular client (my real estate developers, mostly) I try to anticipate their needs for signage, and ask them to help me with their thoughts on their possible future needs. So far, I’ve had very few instances where I had to rescan for the big output. A lot of my stuff is done in Illustrator, so the point is moot.
500 lines/inch sounds rather small as the typical lens/micrograin film can achieve 90 lines/mm. That translates toalmost 2300 lines/inch. I realize that those numbers drop rather fast, but given the precision of enlarging optics I cannot believe that the final resolution is only 500 l/in.
But then this isn’t based on anything but gut! π
If it’s for a regular client
what if it’s personal? do you archive your old photos via scanning?
500 lines/inch sounds rather small as the typical lens/micrograin film can achieve 90 lines/mm.
Talking about 3200 [d|p].p.i. earlier, I was refering to scanning film, not paper.
Bolex movie camera lens were supposed to do 300 line pairs/mm. Of course that’s an aerial image. On even the slowest film, if the wind is behind you and you don’t breathe, and it’s right at the center of the lens, you’ll be lucky to get 120 (a consumer-grade zoom might do 30-48). On film that we are likely to use, there’s probably nothing usable beyond a 4800 p.p.i. scan. And if you’re gonna print a 3×5, there’s no reason to wait for that many pixels.
wrote:
In the referenced "A Few Scanning Tips", Wayne Fulton asserts that color photo prints have a maximum resolution of 200 dpi. To pick up all of 200 dpi, you must scan at 1.5 times 200 dpi (= 300 dpi).
To be clear, the 1.5x – 2.x rule applies to images being printed with a halftone – optimum scanning resolution (ppi) is 1.5 – 2x the linescreen (lpi). Optimum pixel dimensions for photo printers is usually specified by the printer themselves, and I’ve seen it vary from one service to another.
And it’s ppi (pixels per inch), not dpi (dots per inch, which is for output resolution only).
Brian
And it’s ppi (pixels per inch), not dpi (dots per inch, which is for output resolution only).
Webster’s New World:
dpi
abbrev. Comput. dots per inch (a measure of screen or printer resolution)
MS Manual of Style:
dpi = Dots per inch
ppi = Points per inch
Go figure. Lots of people probably don’t care. There are more important battles to fight. Like getting people to understand resolution and scaling.
Websters is wrong by omission on the d.p.i. definition.
MS Manual of Style is outdated on the p.p.i. definition insofar as the way most people should know it these days.
MS Manual of Style is outdated on the p.p.i. definition insofar as the way most people should know it these days.
That’s the point. Look at MS Windows help, and everything is dpi.
It’s a good thing I didn’t quote what they said about resolution and scaling (and color management).
The Minolta 5400 at maximum resolution and bit depth produces files of approximately 233 M. At first, I believed this to be somewhat exaggerated. Somebody pointed out to me that you can burn more than 20 of those on a DVD. If you look at the evolution of file sizes, in a few months or a year or two, this is going to look normal. Scanning takes so long, I do not want to do it twice!
It also depends on your purpose. I have stopped shooting film entirely and am attempting to scan my whole collection of negs and slides (I am a bit more fussy about what I want to keep, mind you).
The point is, unless you are scanning a specific picture for a specific print, do it at maximum everything, otherwise you’ll want to start over in a couple of months.
I always believed that ppi was pixels per inch, by the way, referring to a monitor as opposed to dpi, dots per inch, which refers to a printer. Printers do not behave at all like monitors and will translate 300 ppi into dpi which is not the same at all! The question is, what image resolution (ppi) does your printer need to do its job. The answer is generally believed to be between 240 and 360.
Larry, I don’t see how you can draw any conclusions whatsoever about print resolution from film resolution. Prints and film are different media and you can’t guage the state of the art in one from the state of the art in the other. Decades ago, best resolution of the most advanced film was 100 lines/mm. Whether it is any better today is probably a military secret. Film resolution, not lens resolution, has always been the limitation in film-lens systems. When we are talking about scanning photo PRINTS, rest assured that the lenses in the loop are not the limitation excluding, of course, el cheapos.
excluding, of course, el cheapos.
Ahem! Ah reSEMble, Ah say reSEMble that remark, Suh! π
Yes. Tif files do not lose detail. If you save as JPG, the file is smaller, but more detail is lost every time the file is saved.
Question, if you save a JPEG as a TIFF, does it decompress or simply stop compression?
Michel,
"…Printers do not behave at all like monitors and will translate 300 ppi into dpi which is not the same at all…"
Printers DO create pixels from smaller printer dots arranged in a matrix. The number of dot positions reserved to make up one pixel determines the number of color tones available, i.e., the tone depth. Resolution and tone depth are traded off in creating a printer pixel. A printer dot is not a pixel but a printer pixel is made up of printer dots. It is certainly not improper to refer to a printed pixel.
What SHOULD be improper is to refer to printer pixels per inch (ppi) as dots per inch (dpi). I said SHOULD because dpi has so often been misused that now, by common usage, the term may very well refer to pixels and the reader must figure out from the context what is really meant. I am guilty myself of carelessly interchanging the terms and have done so early in this thread.
"…The question is, what image resolution (ppi) does your printer need to do its job. The answer is generally believed to be between 240 and 360, as far as I can tell for most inkjets…"
240-360 ppi is overkill. Depends on the image. I have some great wall hangings at 100 ppi. Remember also that the scanner resolution needs to be 1.5-2 times the inherent resolution of the object being scanned. You need at least 300 ppi scanning resolution to preserve 200 ppi in the specimen. Thus, if your image is coming directly from a digital camera rather than from scanning an existing image, there is no scanning involved and a 200ppi camera output should be as good as a 300 ppi scan. Look for equivalent quality when inputting either one to your printer.
Question, if you save a JPEG as a TIFF, does it decompress or simply stop compression?
It’ll stop the bleeding. Whatever quality was already lost from JPEG, will still be gone. TIFF compression is loss-less, much like a ZIP archive; the algorithm used doesn’t degrade image quality. The resolution was lost the first time the image was saved as JPEG, and it’s compromised again each time it is reloaded then saved again.
If you have to use JPEG for the master image (it would be nice if you didn’t), always save it in the highest quality setting. When you need a nice, small, JPEG, export a copy under a different name, with all of the compression options you like.
Printed resolution remains a "perpixelating dilema" for me. Although, I have learned a lot from my quest for knowledge—especially from Wayne Fulton’s book "A Few Scanning Tips"–(I purchased his book, which is a VERY valuable tool in my library)as well as following threads like this where experienced people share their view points.
Bottom line for me, is that there are so many variables involved between the quality of the original image that you scan (or import) followed by understanding your scanner’s ability (don’t scan above your scanner’s optimum resolution sort of thing) compounded by the printer you use and then throw in the variable of the quality of the paper—and wala–no two situations are going to be equal to establish the "golden rule" of scanning. Then to further complicate things, if you are scanning for a certain useage and then down the road you want to do something bigger etc., the guidelines change as to what is best.
My words of wisdom are to conduct an experiment for yourself on the same photo– scan in at different resolutions etc…print at different resolutions and I think that you will find as I did—other than the basic concepts that everyone agrees on (you can’t ask for more "pixel power" than what exists)—there are no hard and fast rules that totally apply to every situation.
Patty
don’t scan above your scanner’s optimum resolution sort of thing
optical resolution. not optimal. i.e. you scanner says it can scan to 9600 dpi (don’t say it!) but in the fine print somewhere you see a spec that says 1200 optical resolution. That (1200) is the max you want to scan at (imo). If you go for more you introduce scanner interopolation (up-resing) done by the scanner software. I think PS is better at it. YMMV.
Dave,
optical resolution. not optimal
Oops, I used the wrong "O" word–my bad….
But since I already said this is very "perpixelating" to me—I’m sure all will be forgiven….:)
I exhausted a few brain waves trying to decipher "YMMV"—-
I went from "You make me vomit" to "You make me vroooooooooooom" and I finally settled on "Yes, many, many variables"—
So pal, translation please—
Patty π
George,
If you go back and read what I said, I agreed that the print did have a resolution less than what a good neg can handle.
I did an experiment back in my 8×10 days. I loaded some holders with Brovira, and went out to do some shooting. (ASA of about 1.5 or so, as I remember). The resulting image was phenomenal, in that the same shot on film and then contact printed did not hold as much detail.
I wish I still had that shot!
I would conclude that somewhwere between the neg and the print, even as a contact print, something is lost. What and how is unknown to me.
I would conclude that somewhwere between the neg and the print, even as a contact print, something is lost.
Which is why in process camera work, when you make a contact print, the negative and raw stock are sandwiched together under a vacuum, and the pinpoint light is way up in the attic. Very sharp prints.
I’m sure all will be forgiven..
nothing to forgive! π
YMMV – Your Milage May Vary
Eureka! I found the long-lost article I cited earlier:
Image Scanning, How Much Resolution is Enough?. M.E. Gruchalla, P.E., Member:Editorial Review Board, 4816 Palo Duro, NE, Albuquerque, NM 87110, Communications Quarterly, Summer 1999.
Here is an excerpt pertaining to scanning color prints:
"The resolution limit for a reasonable-quality, bulk processed print is on the order of 10 lp/mm, or about 250 lp/in…We must scan such a print at a resolution of at least 500 dpi(2X) to capture all the useful spatial data in the print image—750 (3X) if we are to reach the contrast reversal resolution, and perhaps 1250 dpi (5X) to retain virtually all spatial information. A typical magazine photo is printed at about 133 dpi, so a scanning resolution slightly greater than about 266 dpi (or about 300 dpi) is needed to capture a printed magazine image…"
"YMMV – Your Milage May Vary "
Thanks for the translation Dave–never would have thought of that one!
(I knew I wasn’t close with the "you make me vomit"—but it did make me laugh that the thought jumped into my head!
Patty
10 lp/mm… 2X…
So, a good piece of film is about 72 lines/mm. We need 72*2=144 pixels per millimeter. So, 144*25.4=3657 pixels per inch. Take the ultimate of 120, and you get (120*2)*25.4=6096, as the absolute most you’ll get out of a piece of film.
r_harvey
Scanning at 2x the lp/in is a practical rule-of-thumb, not to be confused with the "ultimate" and "absolute" modifiers you have used. To squeeze out the last drop of data, however useless, Gruchalla doesn’t throw in the towel until you get to 5x.
(72*5)*25.4=9144
(120*5)*25.4=15240
Maybe I’ll try scanning at 6400 again (hate to only get five of ’em on a CD-R, though).
r_harvey
Since file size goes as the square of the resolution, scanned x5 files are SIX times larger than scanned x2 files. Thus, they take 6 times longer to scan and process. If you’ve got that kind of time, you must still be working! Us retired folk are too darn busy π
Interesting stuff. I’m not suprised at the numbers with which r_harvey came up.
As for contact prints, I didn’t have a vacuum frame but I did devise a solid, tight, hinged back that held the sandwich very close. My illumination was not a true point source; I used a condenser enlarger as the source without film in the carrier, but foccused such that the light, condenser and lens formed a light path similar to a point source. It was pretty sharp. I used enlarging paper, (Brovira, Medalist) not contact (Azo).
The big hassle was the Newton rings. How I hated them!!! At the time I was also working on some new oscilloscope cameras, and we were outsourcing the lens coating to an outfit that developed a good triple coat. I showed them my problem and asked how much they would charge me to do an AR triple coat on a piece of 8×10 glass. I expected a roar of laughter at the idea but they were very interested and offered to do it for a mere pittance. However, the sheer cost of continuing doing 8×10 shots was weighing heavily on family finances, so I backed off to a more economical approach. I never did take them up on it. Pity. I would have had the only contact printer with an AR coated surface.
George,
2x over-sampling is OVERKILL!
The following is much closer to the "trutrh":
I’ve always heard the "rule-of-thumb" number as ranging from 1.5 to 2.0x
thelinescreen frequency of the output device. <<
This little bit of conventional wisdom is so firmly embedded in the consciousness of the industry that I sometimes despair of *ever* rooting it out. But I continue to try. <g>
The easiest place to find quick reference on this is in the January 1994 issue of Pre magazine in a column by Brian Lawler.
http://www.thelawlers.com/FTP/Resolution Essay.pdf.sit) But it has also be covered elsewhere, among other places, in Publish, in the July 1992 issue in an article written by Jane Hundertmark about the process the San Francisco Examiner went through in switching their production over to a digital process.
To keep it simple, it goes like this. When you halftone an image, the *best* resolution you can get is the frequency of the halftone screen. So if you use a 1:1 oversampling ratio, your image resolution, (at 1:1), should be that of the halftone screen. 100, 120, 130, 150 ppi. HOWEVER, this is only going to work really well if the halftone screen is aligned with the pixel grid. If it’s turned at an angle, you’ll get dropouts and really bad averaging. So you need to get between 1.16 and 1.18 pixels per dot as a minimum to get good averaging and to compensate for screen angles. You can improve the quality of output on
any image to an oversamping ratio of about 1.25:1. Problem images, (herringbone tweed, for example), can require as much as 1.4:1. Beyond that, nobody can tell the difference.
But don’t take my word for it. Try it for yourself.
Oh, and BTW, You mentioned that Bruce and David used the "standard" rule of thumb in RWPS3. You’re right, they did. But take a look at the images illustrating concepts in oversampling in Blattner and Roth’s _Real World Scanning and Halftones_. They are resolved at 186ppi. 1.4 times a 133 lpi screen. <very wide grin>
Egads,
Wiz
06-Feb-97 at 18:48:17
Geroge,
I’ll give it a try at 100 ppi. The lowest I ever tried was 240.
The big hassle was the Newton rings.
Which reminded me of the thread where someone wanted to scan glass plates. Old emulsions were quite thick and coarse (you could see the raised images if you looked at them from the right angle). In scanning very old film/plates, there probably wouldn’t be Newton rings. A greater concern with plates would be getting a thin dust bunny grain between the glass and the film–could dig right into the emulsion.
Len,
Thanks for contributing to this discussion. It is probably painful for you to see the same old issues rehashed with previously enunciated truths all but ignored. At the risk of irritating you, however, I am going to rekindle some basic confusions.
The quality (resolution) of the printed image is determined by the weakest link in a 3-link chain: The resolutions of (1)the specimen to be scanned,(2)the scanner, and (3)the printer.
The scanner, if properly set, is rarely the weak link, so quality is generally constrained either by limitations of the original image or of the output device. You seem to have addressed the latter (output) whereas, in this thread, I have been considering the input to be limiting.
You may very well have a printer in mind when you scan but, more generally, you are more likely to scan for usability with other unnamed printers, perhaps of superior quality. So you scan to the limit of the original for an unknown printer presumed not to be the weak link.
Information theory establishes the sampling rate needed to recognize a line-pair pattern as a number higher than the line-pair frequency. The higher the better, of course, but for practical purposes that number is taken to be in the range 1.5 to 2 by most experts in that field.
If you accept the 1.5-2 multiple, all you need to know before setting the optimum scan rate is how refined the original is. Here, per Grochalla for examle and many others, your typical color photographic PRINT (not film) is limited to about 250 lpi.
Len, could it be that the lower scan rates (1.16 to 1.4) you cite apply when it is the printer rather than the scanned specimen that is limiting?
1.5x is hedging. 2x is necessary to avoid aliasing, and to be able to define a sine wave 2x is mandatory.
Basically it comes down to this. The -3db point (the 1/2 power point) is 1/2 the sampling rate in a well behaved system (gaussian). Now, this figure may not be as significant in spatial considerations as in temporal, allowing then for the loosed 1.5x.
At any rate, George beat me to the punch as to what is under consideration.
Well said, George. I don’t even think printer when I scan. Which one? I don’t have a clue at the time, and I don’t know what final size I may want. My 2 1/4’s blow up to 4’Sq very well.
George,
could it be that the lower scan rates …apply when it is the printer
rather than the scanned specimen that is limiting?<<
Well, I don’t like the term ‘limiting’ for starters <g>.
The image rez/ linescreen ratio is, of course, referring to conventional halftone output, since that is the only place that a linescreen will exist. As to original scanning resolution, then that is a different argument……
1) Scanning at the scanner’s optical resolution will produce the the most faithful scan
2) Assuming that image pixel-size corrections need to be made, then any re-sampling is probably better achieved within Photoshop than by the scanner firmware, although that may be dependent upon the scanner quality.
Larry,
"…Now, this figure may not be as significant in spatial considerations as in temporal, allowing then for the loosed 1.5x…"
Gruchalla says it very well as follows:
"Digital photography and image scanning are precisely the same signal-processing process as digitizing an analog electrical signal. In processing typical electrical signals, care must be taken to assure that the sampling frequency is adequately above the highest frequencies in the signal to be digitized."
And later he says
"The digitizing of an image, whether the image of a scene being photographed with a digital camera or a photograph scanned with a scanner, is precisely the same signal-processing process as digitizing a temporal signal. We are simply digitizing in the spatial rather than time domain"
M.E. Gruchalla, by the way, has rather lofty stature in the scientific community, having been called upon to testify before the US Congress to give his views, as a scientist, on what needs to be done to improve our patent laws.
Len,
"…You obviously haven’t bothered to follow the link <
http://www.thelawlers.com/FTP/Resolution> Essay.pdf.sit…"
Although directed to Larry, your remark also applies to me. I was unable to get on that site. However, I tried again just a moment ago and was successful. Looks like a great site and I will definitely go back and check your referenced article. Thanks for the lead.
"…The image rez/linescreen ratio is, of course. referring to conventional halftone output since that is the only place that a linescreen will exist…"
It is not necessary for a linescreen to physically exist to use the concept in defining inherent resolution. If the scene had a grid of alternating black and white lines, and you were able to shrink the grid, at what spacing/thickness would the photo print fail to reveal that such a grid even existed, let alone showing it in sharp contrast. So when the lp/in of a photo print is referenced, the reference is not necessarily to be taken literally. It is a measure of the detail beyond which the original print-creation process is incapable of revealing that any detail exists. And, once again, as I understand it, a typical color photo print of the kind that most of us would be scanning, that lp/in is around 250.
If the scanner and printer to which you feed the scanned output both have infinite rez, the printed output rez is no better than that of the original. It will be worse if you fail to scan at 1.5-2x the original (375-500), and the file will be needlessly large (and cumbersome) if you scan at higher rez.
Len, I have followed the link and it says it cannot be found….
and, George, I was being a bit accomodating about the 1.5 figure being acceptable, as I haven’t actually run any measurements to both quantify and qualify these numbers for spatial considerations. I know quite well that the math is the same and that the change from temporal to spatial is not a governing factor. What does matter is what the final result is to the observer, and there are differences in acceptability between the eye, ear and a state of the art measuring instrument. My comments are from the latter. I worked with samplers starting in 1961, when Tektronix introduced their first sampling plug-in. I wqas one of a small number of individuals in the field offices to specialize in this unit, and later was involved in all aspects of sampling design and production.
As I said, 2x is minimum, in a well behaved system of which there are few when you are state-of-the-art.
Besides, the only reason to even consider all this nonsense is that, if I have to deliver a number of files over the internet, it behooves me to make the files as small as practical, involving trade-offs between compressing a larger file or skirting quality issues by supplying a smaller file to begin with. I just delivered two jobs, one for an annual report and one for a book jacket cover, both at the highest level of quality. I delivered them on CD. No compromises necessary.
The annual report is at 350 dpi for 175 line screen, B&W. I don’t remember the jacket numbers, but I delivered 2x.
And, Len, if you look at the sharpening profiles from PK Sharpeners, they also assume a 2:1 ratio for sharpening for half tone output.
I guess we are all out to lunch. π
Good discussion however.
BTW did you know that sampling predates direct measurement and display of a alternating voltage? When AC was first introduced, a sampling system plotted the waveform.
Larry,
Just go to www.thelawlers.com without the add-ons and navigate from their home page.
George,
The essay appears to have moved (although Brian has many other interesting ones there!).
I’ve e-mailed him to try and get a copy and asked for permission to put it on our site for your (and others) convenience.
Thanks. I found it.
The thermal noise equation is just that, an equation that reconciles Resistance, temperature and bandwidth. Since temp and resistance are not a factor in spatial applications, we only need to consider bandwidth. Also note that the out put is noise voltage squared; that’s because the equation yields it’s answer as a power function, and the equation relating power to voltage is P= E^2/R, and since R is not a factor, one has to look at how this equation actually applies to spatial considerations.
But, that’s not the real problem here. We have considerable confusion arising from the misapplication of terms. Sampling is an unfortunate term to use for deciding what the file size should be. The question really revolves around a ratio in digital terms. Sampling refers to acquisition of data from an ANALOG signal. We are not doing that, and I suspected as much as this conversation progressed. For a really good (but very technical) exposition of all this go to:
<
http://www.geocities.com/bioelectrochemistry/nyquist.htm>
One of the surprises is that Nyquist was first with sampling theory. I have always referred to Shannon. Nyquist, besides the noise in resistors equation, was a name I associated with feedback design.
Another term that raises a red flag is STOCHASTIC, another word for random. This application in spatial terms uses frequency modulation which is correct, but FM is not stochastic. It is periodic, but like any signal processing, FM can broadcast a stochastic signal.
We would call that noise.
Lawler’s paper is fine, and is a good guide, but even as he stated, is only a guide. Look, if I supply a file too large, downsampling is quick and easy. Upsampling is worse, so which side to err?
Larry,
The scanning sensors ARE analog devices. Their responses are digitized. The pixel’s values in each of three color channels are "samples" just as surely as a voltage would be, but the values are quantified (stepped) rather than continuous. That makes them cruder but no less samples. I don’t understand the reservations you expressed.
