On the resolution and sharpness of digital images...

A forum to ask questions, post setups, and generally discuss anything having to do with photomacrography and photomicroscopy.

Moderators: rjlittlefield, ChrisR, Chris S., Pau

rjlittlefield
Site Admin
Posts: 23561
Joined: Tue Aug 01, 2006 8:34 am
Location: Richland, Washington State, USA
Contact:

Post by rjlittlefield »

The way I see it...

8.2 pixels per lp is oversampled. The image should be printed at higher than 300 dpi (for standard viewing conditions). For web display, the image should be resized smaller. At 4 pixels per lp, the resized version will contain all the detail of the original, but will still look a bit fuzzy. At 2 pixels per lp, the resized version will look quite sharp, but will probably be missing some detail. 3 pixels per lp might be a good compromise.

3.25 pixels per lp is what I might consider optimally sampled, meaning that "good compromise" I spoke of. Given that resizing always screws things up at least a little, I'd be inclined to leave the image original size.

4.3 pixels per lp is a bit oversampled. Printing at somewhat higher than 300 dpi would be reasonable. So would resizing to 3 or even 2 pixels per lp, depending on how you feel about that tradeoff between sharpness and lost detail.

In all cases, sharpening the resized images with a narrow filter (e.g. USM with radius 0.5-0.7 pixels) can help to bring out detail that might otherwise be overlooked due to low contrast. (The idea is to "add some snap", without introducing visible artifacts.)

--Rik

Charles Krebs
Posts: 5865
Joined: Tue Aug 01, 2006 8:02 pm
Location: Issaquah, WA USA
Contact:

Post by Charles Krebs »

Rik,
sometimes a bit counter-intuitive isn't it...

BTW, these are two links I have found useful in this area:

http://micro.magnet.fsu.edu/primer/digi ... asics.html
http://www.microscopyu.com/tutorials/ja ... index.html

Charlie

rjlittlefield
Site Admin
Posts: 23561
Joined: Tue Aug 01, 2006 8:34 am
Location: Richland, Washington State, USA
Contact:

Post by rjlittlefield »

I'm curious -- which aspects do you find counter-intuitive?

--Rik

Epidic
Posts: 137
Joined: Fri Aug 04, 2006 10:06 pm
Location: Maine

Post by Epidic »

Rik, the 300dpi number is based on an 8x10 print at the correct viewing distance which is equal to the diagonal of the print. A 16x20 would only require 150 dpi. Or 600 dpi for a 4x5 print. Which means that, like film, there is no print size limitation to a digital image. Fixing print resolution to an absolute number makes no sense. That is not to say you cannot use different resolutions, but the 300 dpi falls under the permissible circle of confusion criteria. Just like you can change that critera for DOF, you can change it for acceptable sharpness - it is subjective.

However, the reason the 300 dpi number has been used by itself without regard to print size or viewing distance is, for want of a better word, laziness. When I confronted the engineers at my old company (a Japanese digital camera manufacturer) why they didn't base it on viewing distance / print size, he said, while I was right, it was just easier to use a fixed print resolution. Certainly no one was complaining.
Will

rjlittlefield
Site Admin
Posts: 23561
Joined: Tue Aug 01, 2006 8:34 am
Location: Richland, Washington State, USA
Contact:

Post by rjlittlefield »

Here is a rework of my earlier illustration, this time starting with a smooth wave for intensity, instead of a hard-edged set of bars. I also added a line at 8.2 pixels per line pair, to cover the range that Charlie asked about.

Image

--Rik

Charles Krebs
Posts: 5865
Joined: Tue Aug 01, 2006 8:02 pm
Location: Issaquah, WA USA
Contact:

Post by Charles Krebs »

I'm curious -- which aspects do you find counter-intuitive?
That you would obtain a "sharper" looking result by taking fewer samples.

At least for me, it would seem intuitive that if more samples taken of any subject, the more accurate the reproduction could be, and that "non-sharpness" would decrease with greater "over-sampling".

rjlittlefield
Site Admin
Posts: 23561
Joined: Tue Aug 01, 2006 8:34 am
Location: Richland, Washington State, USA
Contact:

Post by rjlittlefield »

Epidic wrote:However, the reason the 300 dpi number has been used by itself without regard to print size or viewing distance is, for want of a better word, laziness.
Could be, I don't have the data. I use it as shorthand for viewing at 12 inches etc.
Epidic wrote:...the 300dpi number is based on an 8x10 print at the correct viewing distance which is equal to the diagonal of the print. A 16x20 would only require 150 dpi. Or 600 dpi for a 4x5 print. Which means that, like film, there is no print size limitation to a digital image.
I think we're agreed about the conditions where 300 dpi is the right number -- at least if you're willing to consider diagonal(8,10) = 12.8 to be "around 12 inches", as I wrote.

About the rest, I'll disagree. There's not much reason to think that a 16x20 print is only going to be viewed at 25.6 inches (or more), and viewing a 4x5 print at 6.4 inches is highly unlikely because most people's eyes won't even focus that close. I will happily agree that 150 dpi for the 16x20 and 600 dpi for the 4x5 is what's required to pack the same information as 300 dpi on an 8x10. But what's required to look sharp is not the same as what's required to pack the same info.

Here's an experiment to try. Print the same digital image at all three of those resolutions and sizes: 600 dpi 4x5, 300 dpi 8x10, and 150 dpi 16x20. Then hand them all to volunteers sitting at a well-lighted table, and ask them which looks the most sharp.

If I understand your theory correctly, they should respond by holding the 4x5 print at 6.4 inches from their eye, the 8x10 at 12.8 inches, and the 16x20 at 25.6 inches, then saying "Gee, they all look the same to me."

When I run the experiment on myself, that's not even close to what happens.

I view the 8x10 at around 12 inches, which happens to be equal to its diagonal and also happens to be almost as close as I can focus comfortably. I view the 4x5 at around 10 inches, which really is my closest focusing distance. And I view the 16x20 at around 16 inches, that being a comfortable holding distance where I can still see the whole thing easily, perhaps by moving it around a bit. (I simply can't reach 25.6 inches -- my arms are several inches too short!) These distances are not the same as the diagonal of the print, but they are what I do.

So my response, under what I consider to be perfectly reasonable viewing conditions, is to say that the 4x5 looks very sharp, but it's not big enough for me to really tell how much detail is in it. The 8x10 looks both detailed and sharp -- an excellent print. And the 16x20? "Well, it's big, but it's not really sharp -- I guess maybe it would look OK if I backed off, but dang, then it's not really a sharp 16x20, it's just a sharp 8x10 blown up with empty magnification."

Looking at the issue a different way, what's the justification for saying that the "correct" viewing distance for a print is equal to the print's diagonal?

Max Lyons reports that when confronted with a 300 dpi wall-sized print of his giga-pixel Bryce Canyon, what people actually did was to keep moving back and forth and sideways -- varying their viewing distance from, oh say, a wall's diagonal away, to about a foot.

They did that because they were free to and because there was detail at all scales down to as small as they could see, pulling them in to see more.

To borrow some of your words, defining required resolution as a fixed fraction of the print diagonal does not make much sense, as it seems to be an arbitrary aspect with no relation to how people behave when you give them the opportunity.

Of course if you don't let the viewers look close, then finer detail doesn't matter. And if there's no finer detail to see, even when they do look close, then they'll soon stop bothering to do that.

I guess it makes the analysis easier, but it seems, for want of a better word, simplistic. :wink:

--Rik
Last edited by rjlittlefield on Thu May 10, 2007 9:23 pm, edited 1 time in total.

rjlittlefield
Site Admin
Posts: 23561
Joined: Tue Aug 01, 2006 8:34 am
Location: Richland, Washington State, USA
Contact:

Post by rjlittlefield »

Charles Krebs wrote:
rjlittlefield wrote:I'm curious -- which aspects do you find counter-intuitive?
That you would obtain a "sharper" looking result by taking fewer samples.
Expressed like that, the statement seems counterintuitive to me too.

But I think that's because the words hide so much of what's going on.

It's not the taking of fewer samples, per se, that increases apparent sharpness. The increased sharpness comes on the display side, from compressing the same amount of information (the same number of cycles) into fewer arc-minutes of visual field.

Given no other constraints, the best strategy is to sample zillions of pixels and then display them at pixel size that puts the smallest detail of the image close to the resolution of the viewer's eye. Oversampling is great under these conditions.

But if your pixel size is fixed and big (computer monitor), then your strategy has to change. In that case, you want to either sample or resample (resize) so that the smallest detail of the image covers only enough pixels to make it visible.

All of this presumes, of course, that your definition of "sharpness" agrees with mine, which means that it's something like "amount of discernable detail per degree of visual field", under whatever viewing conditions you do the evaluation. (See discussion above, in my response to Will, for more about that point.)
At least for me, it would seem intuitive that if more samples taken of any subject, the more accurate the reproduction could be, and that "non-sharpness" would decrease with greater "over-sampling".
These words sounds like equating "accuracy of reproduction" with "sharpness". That seems reasonable as words, and it makes sense as images too, if the thing being reproduced is itself sharp when displayed at the final size. If the thing being reproduced would be fuzzy at final size, then it has the dreaded "empty magnification" and you can make it look sharper (my definition) by also making it smaller. If you're printing, you may be able to use smaller pixels; for a monitor, you need to use fewer of them.

Does this explanation match your subjective reaction to my illustrations?

If yes, that would be nice to know, and if not, I'd like to see if we can figure out how to resolve the discrepancy.

--Rik

Charles Krebs
Posts: 5865
Joined: Tue Aug 01, 2006 8:02 pm
Location: Issaquah, WA USA
Contact:

Post by Charles Krebs »

Rik,
Given no other constraints, the best strategy is to sample zillions of pixels and then display them at pixel size that puts the smallest detail of the image close to the resolution of the viewer's eye. Oversampling is great under these conditions.
See... I find that quite "intuitive". :wink:

I have had similar discussions elsewhere where "oversampling" was treated like the plague. Problem is, as I alluded to earlier, it's near impossible to avoid either over or under sampling if you use a variety of objectives unless you are constantly changing you hardware setup. So my approach has been to set things up so they will properly accommodate my most "demanding" commonly used objective(s). This would be my 4/0.16 and the 10/0.40 used with a 1.67X photo-eyepiece. With the 20x, 40x and 100x things will then be progressively more "over-sampled". I see this as a preferable situation to setting up for the 40X or 100X and then getting considerable under-sampling with the lower power objectives... you can't make something out of nothing!

Charlie

rjlittlefield
Site Admin
Posts: 23561
Joined: Tue Aug 01, 2006 8:34 am
Location: Richland, Washington State, USA
Contact:

Post by rjlittlefield »

Charles Krebs wrote:I have had similar discussions elsewhere where "oversampling" was treated like the plague.
Interesting. I don't understand why oversampling would be a plague unless it meant that you were missing an opportunity to capture a wider field, or triggering some other unacceptable cost like too much storage or bandwidth or computation. I suspect that people who decry oversampling developed that mindset in a situation where it really did trigger such problems, and have not stopped to consider that other situations have different tradeoffs & constraints.
I see this as a preferable situation to setting up for the 40X or 100X and then getting considerable under-sampling with the lower power objectives... you can't make something out of nothing!
I agree. And the fact that you told me earlier what fraction of the eyepiece field you're capturing, also tells me that you've thought about that tradeoff. So it sounds to me like you have everything nailed. No surprise there! :D

--Rik

Epidic
Posts: 137
Joined: Fri Aug 04, 2006 10:06 pm
Location: Maine

Post by Epidic »

Rik, viewing distance is a yard stick - this is a subjective measurement. The idea is that to see the entire image area, you simply view it from a relative point which happens to be about equal to the diagonal of the area - I think the angular field is around 50 degrees. Yes, individuals look at prints at different distances. Go into a gallery and most stand back. Go to a GigaPixel exhibition where they challenge the viewer to find detail, then they go very close - to close to see the entire image. (The GigaPixel project is not really about making good images, just really detailed ones to WOW the audience.)

BTW, this is also related to depth of field. Depth of field changes with viewing distance. Which means sharpness changes with viewing distance.

Certainly you can have a 16x20 print at 600 dpi. You have monitor images at 72 dpi. My avatar looks good simply because the viewing distance relative to the image size is so long. However, the correct viewing distance idea is to set a clear starndard in relation to the human visual system in order to set a reference point. Without a reference, a simple absolute linear resolution will make no sense. You can actually change the criteria for the correct viewing distance just as it is done with depth of field, but there is a criteria.
Will

Epidic
Posts: 137
Joined: Fri Aug 04, 2006 10:06 pm
Location: Maine

Post by Epidic »

I'm glad sampling has been brought up. It is a strange idea. What difference does pixel pitch make? Pixels by difinition should not be resolved by the viewer.

Back in the old days of film, no one worried about sampling when they switched from 100ISO film to 400ISO film. My 135mm Optar has less resolving power than my 35mm Hexanon and even when shooting on the same emulsion, my 4x5s from the Optar outshines the 35mm images from the hexanon. Why does a large-format image do so well? The "sample rate" is no different than 35mm with the same emulsion.

When we matched optics to a sensor, what was more important was the sensor size, not the pixel pitch. The sharpness/DOF criteria for a 6mp APS-C sensor was the same for the 12mp APS-C size sensor. The pixel pitch was not discounted - the pitch impacted the MTF limit.

Now Rik suggest this is not important. Why is this not important? Sharpness and detail are qualities percieved in the viewer, not contained in the image in an absolute sense.
Will

rjlittlefield
Site Admin
Posts: 23561
Joined: Tue Aug 01, 2006 8:34 am
Location: Richland, Washington State, USA
Contact:

Post by rjlittlefield »

Epidic wrote:Now Rik suggest this is not important. Why is this not important?
Please clarify. What it is that you think I have said is not important?

--Rik

Epidic
Posts: 137
Joined: Fri Aug 04, 2006 10:06 pm
Location: Maine

Post by Epidic »

I'm sorry if I misinterpreted your comments when you answered my first posts. I believe you said this had nothing to do with what you were saying. I believe it cannot be ignored.
Will

rjlittlefield
Site Admin
Posts: 23561
Joined: Tue Aug 01, 2006 8:34 am
Location: Richland, Washington State, USA
Contact:

Post by rjlittlefield »

Epidic wrote:I'm sorry if I misinterpreted your comments when you answered my first posts. I believe you said this had nothing to do with what you were saying. I believe it cannot be ignored.
Will,

I'm sorry too. I did not mean to casually dismiss anything, and I tried to explain the rationale for approaching the analysis as I did. Obviously the explanations didn't work -- that happens sometimes.

I am still not absolutely sure what you're talking about here when you write "this" and "it".

However, I have re-read the earlier posts, and I see that I used the word "irrelevant" twice. One of those was regarding circle of confusion, the other was regarding the effect of contrast on resolution. I wrote at some length about contrast and resolution, so I speculate that "this" and "it", here, refer to circle of confusion.

"Irrelevant" was not a good word for me to use regarding circle of confusion. Most everything eventually traces down to circle of confusion. It represents the smallest detail the viewer can resolve, so it is extremely relevant to all the phenomena that I am interested in.

The difficulty is that in relating everything back to the value of a circle of confusion, as I perhaps mistakenly thought you wanted, the discussion can quickly get more complicated than it needs to be.

In the sentence following "irrelevant", I wrote that "I'm presuming that the human either has or can get enough magnification to examine the individual pixels if necessary". Once I've said that, then for my purposes the circle of confusion has become sufficiently small that it's no longer a limit. The ability to discern detail, in that case, is imposed by pixels per cycle, or if you like, samples per cycle. I should have added, explicitly, that in order to interpret my illustrations as I intended, the viewer must sit close enough to a good monitor that the individual pixels can be discerned. That aspect slipped my mind because it's the usual case for my eyes and my monitors.

When the discussion shifted over to printing, a second and very different situation arose. In that case, pixels are typically at or smaller than the circle of confusion.

When the pixels are very small, the analysis is easy -- one just expresses everything in terms of the circle of confusion and scales appropriately. However, the circle of confusion then appears in both the numerator and denominator of most expressions I care about, so its actual value is not a particularly useful thing to pay attention to ("is irrelevant"). The more natural and convenient parameter, IMO, is actually circles of confusion per cycle, and so on, and the natural limit is to "display [the images] at pixel size that puts the smallest detail of the image close to the resolution of the viewer's eye", as I replied to Charlie.

When pixels are very nearly the same size as the circle of confusion, as with 3000 pixels across a 10" print, then life gets more complicated. In that case, the perceptual effects are due to a combination of digital sampling and visual circle of confusion. Quite frankly, I think this regime is not well understood. If it were, then we would not have some people confidently asserting that 300 dpi is sufficient at 12" (industry norm), while other people say "but 600 dpi is better" (still at 12"), and at least one person (Roger N. Clark) says that 600 dpi is needed even at 20" -- and all of them can proffer experimental evidence to support their version of the truth.

Perhaps this discussion is a step toward meeting your request to "just relate this whole experiment to permissible circles of confusion". In any event, I'd appreciate your re-reading the thread with these viewpoints & interpretations in mind, and see if it strikes you as less contentious.

Thanks for prompting me to try clarifying these points.

--Rik

Post Reply Previous topicNext topic