Impact of DPI on scan quality

I have a lifetime's worth of Kodachromes, Ektachromes, and sundry negatives that I would like to digitize and print. Most were taken with either Leica R's or M's and I would like to keep their quality. I've seen that Nikon has announced a new series of scanners with the lower priced scanning at 2900 DPI and the more expensive at 4000. Given that my hoped for output will not exceed 11 by 14 prints - is it necessary to step up to the 4000 DPI unit?

<p>

Thanks for your advice.

i would say yes....

<p>

ive been printing some of my negs scanned at 2700 dpi, with an output

print of around a 9 or 8 inch height or width...this roughly

translates to and output dpi of 300 or so...

<p>

printing a 10 inch pic scanned from a 4000 dpi scanner will yield a

400 dpi image, which is fairly good, but might show some pixelation.

<p>

it really depends on what you are going to do with the prints and the

quality you wish to have. it would be worth getting darkroom prints

for the best quality.

Seth:

<p>

My two cents worth... A grain of silver in film is just under

1/10,000th of an inch in diameter; or stated another way, the

resolution of film is slightly better than 10,000 DPI. I think the

closer your initial scan resolution is to the film itself, the

happier you'll be with the results you obtain in output.

I've been doing all my printing with film scanner and inkjet printer

for at least 6 years, and 80% of it that way for 5 years before that.

<p>

You can make beautiful prints with output dpi of 250 on a 1440 dpi

inkjet printer. At 2700 ppi full frame scan, that's about an 11x17"

print. Certainly a 4000 ppi scan will give you more data to work with,

but I've found that it's only really necessary if you're making

significantly larger prints.

<p>

Godfrey

Also, you should take a look at third party drivers for these

scanners. I use the LS2000 which is rated at 2700 dpi, but if I want

a really hi res scan I go to the silverfast driver which will give

twice the resolution - 5400. The results are amazing seen on screen.

On the other hand I don't print.

<p>

Theoretically this will give you a nearly 18" high print at 300 dpi

full frame. Anything that size is going to be on the wall, i would

have thought, so you won't be looking close to see the grain anyway.

<p>

There are most likely silverfast drivers for both these new scanners

and they will probably be bundled with them, so that may solve your

problem.

<p>

Rob.

Since it's a hypothetical question (i.e., at this point you don't

really know what you're going to do with the scans, exactly) the

appropriate hypothetical answer in this case would be "more is better"

:-) You can always choose to drop some resolution, but you won't be

able to get more if you need it.

I think that more resolution is always better. I have the Minolta Dual

II (2780 dpi) and it gives excellent results up to 8 x 10 on a Epson

870 @300dpi. For color you are hard pressed to tell much difference

between this and a conventional pro lab print. If you want to go

bigger you need more resolution. Also one of the nice things about

digital is the way you can very easily crop and you tend to do this

more on screen than I used to do in the darkroom. As a result you are

often wanting to blow up smaller areas to larger sizes -then you need

the resolution. I think if you are seriously going to produce a lot of

11x14 prints then you need a 4000dpi scanner, if not and 8 x 10 is

more your typical size then the lower res scanner will work just fine.

Of course I have not said anything about Dmax.....

Call some pro labs and ask what output resolution they use for digital

prints. I'd be surprised if you found 300 dpi to be less than

professional quality.

The Fujix 3000 printer I sometimes use prefers 400 dpi and will take

320 if I insist. There's no 300 input, though. My Epson wants 360.

Some people say above this gives even better results, and some deny

it. . . . .

I see little difference between printing at 500 dpi and 300 dpi on an

Epson 870. Usually I just print at the resolution it gives with out

any resampling, so it usually works out at higher than 300 as I rarely

print an 8 x 10 on the printer. I will hedge my bets and say that the

limiting factor in my opinion is largely the printer not the scanner

and the print quality is not as good as a conventional print at close

distances @ 3 inches and looking carefully. There are indeed scanning

issues (retention of shadow detail etc. etc.), but there are many

issues with printing in the darkroom too that can minimize your

chances of getting the best possible interpretation of your image. If

I had a b & w darkroom I would prefer to use that for b & w, but I

don't, so 35mm digital is a close second. Sometimes coming second is

not good enough - it all depends on you and your time. For 6 x 6 I

still use a conventional darkroom and love the unquestioned high

quality. For color I think if you want to be in control then digital

is 100% easier than a conventional process.

Beyond 4000 pixels per inch,for Kodachrome, you are typically scanning the textures of the film --dye clouds, grain, etc.--, not getting more "image" or "information"

That's about a 5300x3520 pixel image for 35mm.

Going to mythical resolutions like 9200 or 16,400 just make a mess.

You can of course, get images of the real world at much higher resolutions if you want, If, however, you are starting with film, film sets its own limits

.

This is a discussion from 2001, but then again the question never goes away.

Shutterfly, where I sometimes make larger prints (especially on sale) says:

Shutterfly Help Center

That is, 2000x1600 is enough for prints up to 20x30 inches.

Note that it is common to view larger prints from farther away.

Also, I suspect, that a good 2700 dpi scan is better than a

poor 4000 dpi scan. (And there are a lot of poor scanners

out there.)

Dye clouds in Kodachrome are different than most other films,

so it isn't so easy to explain.

The MTF graphs:

https://125px.com/docs/film/kodak/e55-2009_06.pdf

for three Kodachromes go out to about 80 cycles/mm

down to 10%, which is probably farther than you should

really go. (And the three graphs look amazingly similar.)

In any case, 80 cycles/mm is about 4000 dpi.

2700 dpi, 54 cycles/mm, or about 30% on the MTF graph,

is probably more realistic.

At some point, you get grain aliasing, which is worse than

just not enough resolution.

Since this has been resurrected: Can we please differentiate, once and for all, that

DPI stands for Dots Per Inch, and is only relevant and applicable to printers.

While PPI is Pixels Per Inch, and is what we should refer to the resolution of scanners in.

(And can someone please explain that obvious difference to Epson and some other scanner makers)

Pedantic? Maybe, but it helps to clarify whether we're referring to a digital image or to printed 'resolution'.

- Not 'resolution' really, but that's for another pedantic discussion. ;)

The MTF graphs:

 

https://125px.com/docs/film/kodak/e55-2009_06.pdf

 

for three Kodachromes go out to about 80 cycles/mm

down to 10%, which is probably farther than you should

really go. (And the three graphs look amazingly similar.)

In any case, 80 cycles/mm is about 4000 dpi.

 

2700 dpi, 54 cycles/mm, or about 30% on the MTF graph,

is probably more realistic.

Well, at least Kodak aren't claiming ludicrously high resolutions like 300 cycles/mm for film. Unlike Zeiss used to.

Since this has been resurrected: Can we please differentiate, once and for all, that

DPI stands for Dots Per Inch, and is only relevant and applicable to printers.

While PPI is Pixels Per Inch, and is what we should refer to the resolution of scanners in.

(And can someone please explain that obvious difference to Epson and some other scanner makers)

Pedantic? Maybe, but it helps to clarify whether we're referring to a digital image or to printed 'resolution'.

 

- Not 'resolution' really, but that's for another pedantic discussion. ;)

 

Well, at least Kodak aren't claiming ludicrously high resolutions like 300 cycles/mm for film. Unlike Zeiss used to.

Since to OP asked about 11x14 prints, and used DPI in the question, I continued the theme.

Scanned images on disk are fine, but they look much nicer printed.

(And ignoring for now, the resampling done by printers, which may or may not

be done as well as it should be.)

If you extrapolate the Kodak curves down to 1%, they get to about 150 cycles/mm.

I suspect they get to 0% for 300 cycles/mm. (That is, asymptotically.)

Right.

Here's the difference between pixels and printed dots at a high magnification.

Pixels - at several hundred percent view:

Dots - inkjet print at about same magnification:

The inkjet print appears to me to have slightly better resolution, due, I suspect, to some interpolation in the driver software.

And here's the whole frame, taken with a 13 Mp DSLR (4368 x 2912 pixels, or approximately 3080 pixels per inch).

More than adequate for a straight 11" x 14" print, and with modern 'up rezzing' software a mural sized print would probably look great.... but not from that boring shot of an urban back-street above obviously.

This is the sort of nonsense you get by conflating DPI and PPI -

My two cents worth... A grain of silver in film is just under

1/10,000th of an inch in diameter; or stated another way, the

resolution of film is slightly better than 10,000 DPI.

Except that a single pixel can show 255 levels of tone or 16.7 million colors. Something that a single silver 'grain', dot of ink or dye-cloud obviously cannot.

Except that a single pixel can show 255 levels of tone or 16.7 million colors.

Actually no. It can show 16.7 million device values. A device value may not be a color. Two device values can be the same color.

Don't mix up device values with color; just because you can encode something into 16.7 million numbers, doesn't mean they are all colors or differing colors. That 16.7 million value (of which no humans can see) isn't based on colors. It's based on encoding of numbers. Big difference.

http://digitaldog.net/files/ColorNumbersColorGamut.pdf

Actually no. It can show 16.7 million device values. A device value may not be a color. Two device values can be the same color.

Don't mix up device values with color; just because you can encode something into 16.7 million numbers, doesn't mean they are all colors or differing colors. That 16.7 million value (of which no humans can see) isn't based on colors. It's based on encoding of numbers. Big difference.

http://digitaldog.net/files/ColorNumbersColorGamut.pdf

Well, for one, if you wanted to more accurately encode them, they would be logarithmic.

On a linear scale, the difference between 1 and 2 is big, but between 254 and 255 is small.

Well, for one, if you wanted to more accurately encode them, they would be logarithmic.

On a linear scale, the difference between 1 and 2 is big, but between 254 and 255 is small.

The point is encoded numbers are not necessarily colors. G255/R0/B0 in ProPhoto RGB is an 8-bit number and not a color.

Well, for one, if you wanted to more accurately encode them, they would be logarithmic.

They are logarithmic in most colour spaces. Apart from a linear region close to zero.

N^1/2.4 for sRGB

N^1/2.2 for AdobeRGB

N^1/1.8 for Prophoto RGB

N^1/3 for ECI V2.0 RGB

Where N is the normalised fraction of maximum linear pixel value.

Of course, it would be better if the logarithmic function was applied before digitisation, but we have what we're given.

Anyway the point was: Whatever number of colours a device can show or the human eye can differentiate, it's a lot more than the two that can be shown by the presence or absence of a dye cloud or ink dot.

On a linear scale, the difference between 1 and 2 is big...

And the difference between 0 and 1 is infinite!

Anyway the point was: Whatever number of colours a device can show or the human eye can differentiate, it's a lot more than the two that can be shown by the presence or absence of a dye cloud or ink dot.

The point should also be: there are lots of encoded numbers that are not colors. The point should be, there can be millions of encoded numbers that define the same colors. The 16.7 value stated has no direct relationship to colors. Just device values; not colors.

Color, is a perceptual property of in this discussion, humans. Devices or device values or encoded numbers no. So if you can't see it it's not a color. Color is not a particular wavelength of light. It is a cognitive perception, the excitation of photoreceptors followed by retinal processing and ending in the our visual cortex, within our brains. As such, colors are defined based on perceptual experiments.

Fairchild's "Color Appearance Models". Page 1:

"Like beauty, color is in the eye of the beholder. For as long as human scientific inquiry has been recorded, the nature of color perception has been a topic of great interest. Despite tremendous evolution of technology,fundamental issues of color perception remain unanswered. Many scientific attempts to explain color rely purely on the physical nature of light and objects. However, without the human observer, there is no color".

Further on the same page:

"It is common to say that certain wavelengths of light, or certain objects are a given color. This is an attempt to relegate color to the purely physical domain. It is more correct to state those stimuli are perceived to be a certain color when viewed under specific conditions".

Page 1 paragraph 2 of Digital Color Management by Giorgianni and Madden:

"But color itself is a perception and perceptions only exist in the mind".

Page 11 of The GATF Practical guide to Color Management:

"Although extensive research has been conducted, we still not completely understand what happens in the brain when we "see" color. The visual sensation known as color occurs when light excites photoreceptors in the eye called cone cells".

Page 75 of Understanding Color Management by Sharma:

"Color is an impression that we form in our brains".

And the difference between 0 and 1 is infinite!

The ratio is infinite (or zero), but the difference is one. That difference is imperceptible even on a lowly 8-bit scale. Humans perceive differences, not ratios, and on a logarithmic scale at that.

For *some*, to be ignored....

Color and Light – Linear and Log >> Human vs Video

Humans perceive differences, not ratios, and on a logarithmic scale at that.

Aren't differences on a logarithmic scale ratios?