Pixel Size / Resolution

No need.... I was just commenting on "suitable" in context of

So what is the minimum number of pixels on a 35mm sensor that will render a suitable picture? What about maximum

bohdank wrote in post #13438202
No need.... I was just commenting on "suitable" in context of

So what is the minimum number of pixels on a 35mm sensor that will render a suitable picture? What about maximum

Well looking back is one way of finding out the minimum number of pixels needed to render a " suitable picture " depending on what is suitable for the person who wants to know.

You have a point but there is a way to be objective which is what my answer was. Suitable is a moving target.

First off - very few commercially available lenses can out-resolve a modern sensor... so the question should really refer to the lens.

From personal measurements using my ccd camera (4.7 micron pixels, no AA filter) good lenses can achieve something like 20 micron spots on axis... so over a ff sensor you don't actually need more than 50x36x50x24x4 = 8.6 million pixels on the sensor. Now I may be out by a factor of two here... so truly exceptional lenses may need 17 million pixels on a full frame sensor. Increasing the sensor resolution does not provide you with more information - lens aberrations mean that higher resolution information disappears before it reaches the sensor plane.

For comparison - my estimate of lens sharpness agrees with measurements quoted in a recent LensRentals sale. The lenses on offer typically achieve 18-22 line pairs per mm, or equivalently the spot size is around 25 microns.

It's worth remembering that to achieve ultra high resolution, the lens needs to be at its optimal f-stop, technique must be perfect and it must be close to critically focused - it happens but it's not something you can expect from every shot.

So - a sensor that produces better than 8.6 million pixels is probably quite adequate, and a sensor that produces more than 17.2 million pixels almost certainly is.

Since it's been raised again, we should look at the 300 dpi myth. It comes up time after time.

300 dpi corresponds to something like 12 dots per millimetre.

The fortunate few can maybe resolve quarter of a milliradian. (People with normal vision can resolve about half a milliradian.) This means that they could resolve 300 dpi from a distance of about 340 mm (14"). Folk with more regular vision need to go to half this distance assuming they can actually get their eyes to focus.

In reality though, we don't look at 8x10 images from about 14". Ok - you may but most of my prints are viewed from a distance of 1-2 metres. This means that someone with supervision viewing my prints could only actually resolve 50-100 pixels per inch. This time, optical aberrations in the eye limit what you can perceive.

That doesn't mean you should not print at 300 dpi - this is necessary to preserve image smoothness but really you do not need to fill it with invisible information.

Understanding this allows you to print a 5D2 image at about 2m x 3m scale. Provided the viewer is at a sensible distance to view the image, they can't see any loss of resolution.

This business about minimum viewing distance.... go to a gallery. Any prints that suggest detail such as landscape etc. people will view from a distance, be drawn to the image and will view it from inches away. Unless you can fence off your images, I really do not accept the "normal viewing" argument.

bohdank wrote in post #13438396
This business about minimum viewing distance.... go to a gallery. Any prints that suggest detail such as landscape etc. people will view from a distance, be drawn to the image and will view it from inches away. Unless you can fence off your images, I really do not accept the "normal viewing" argument.

Agreed! You don't see too many people going close to examine the detail in a large portrait print.

I think the 300 ppi was selected to fit in to the DoF system based on a practical situation, a typically 8x10 viewed from, well, around 14" - quelle surprise

Even though I look at them quite closely, I often print only with 240 ppi for my landscapes, after appropriate sharpening, of course.

My lab insists on 300 for smaller prints and 240 for LARGER prints, 16 x 20 and up. They basically said, the software that comes with their printers doesn't do a great job at resizing.

I stopped printing at home over a year ago. I don't print often but when I do it's in large batches and couldn't be bothered doing it home, anymore.

bohdank wrote in post #13438604
My lab insists on 300 for smaller prints and 240 for LARGER prints, 16 x 20 and up. They basically said, the software that comes with their printers doesn't do a great job at resizing.

I stopped printing at home over a year ago. I don't print often but when I do it's in large batches and couldn't be bothered doing it home, anymore.

good lucky meeting that requirement at 20 x 30. Took many people purport the 300dpi "requirement" when it really is about 2 x overstated.

Life was so much easier when all we had was film ... it really was!

Sdiver2489 wrote in post #13438821
good lucky meeting that requirement at 20 x 30. Took many people purport the 300dpi "requirement" when it really is about 2 x overstated.

Well they don't have to be real pixels and it's not 2x times, imo. Read my previous posts for my opinion. Anything less (before resizing) and you are losing resolution. If it is "suitable" for ones use, then fine.

As far as meeting that requirement..... ya, well, resize/interpolate. If you don't, the printer will and there is no way to override that.

harcosparky wrote in post #13439571
Life was so much easier when all we had was film ... it really was!

I never printed color film, myself (back when). I was at the mercy of the lab to interpret how I wanted my prints. Hell, not even cropping, unless you used a custom lab and paid the price.

It's still just as simple. Shoot JPG, send to to lab

JeffreyG wrote in post #13438104
Bigger pixels are better so long as the sensor is bigger too (and provided we have the focal length we need and are not cropping).

Imagine a 1" x 1" sensor with 10 million pixels vs. a 2" x 2" sensor with 10 million pixels. Both give the same resolution in print, but the latter sensor has pixels that are four times the size and so it will give four times the light gathering abilty.

Now imagine a 1" x 1" sensor with 10 million pixels vs. a 1" x 1" sensor with 40 million pixels. Now the latter sensor offers twice the resolution with the exact same light gathering area. And while each pixel in the second sensor is only 1/4 the area of the pixels in the first sensor, the higher S/N ratio of the individual pixels is averaged over more pixles per area and so noise is a wash.

To sum up = higher pixel densities on equal size sensors yields higher resolution with no increase in noise. Same pixel densities on larger sensors yields equal resolution with less noise.

The analogy I use is comparing digital to film. Take two film cameras both loaded with Kodachrome (compare to mp resolution). One camera is a 35mm and the other is a 6x7 medium format (larger sensor/larger pixels). Thus the smaller sensor has to work harder for the same size printed output. Not exact science but you get the idea.

noisejammer wrote in post #13438337
First off - very few commercially available lenses can out-resolve a modern sensor... so the question should really refer to the lens.

The above statement is not entirely correct, actually.

It depends entirely on the frame of reference, and 50% MTF, f.e., is not necessarily a good frame of reference for a real photographic subject.

See highlighted piece: weirdly enough I get the same number, but for different reasons, and with a much smaller spot, namely equal to the sensel in size. See below for a sample calculation.

See my above statement.

World class lenses, amongst which are most if not all L lenses, are diffraction limited at F/4, which equates to 400 lp/m at the Rayleigh criterion, 9% MTF, ~512 nm wavelenght (average), which is what we referred to for film. Good lenses 30 years ago managed 250 lp/mm, so where does the low number you mention originate from?

I think you are really talking about system resoilution here, which is equal to the inverse of the sum of the inverse of the lens resolution and the inverse sensor resolution, in lp/mm. The theoretical maximum obviously is only attainable under a very strict regime involving > 10 second MLU before firing the shutter, and and very heavy tripods with remote shutter releases.

I find the number quoted to be rather low, as studies have shown that amateurs with film slrs used to get about 20-40 lp/mm from their colour negatives, and pros up to 60, maybe 80 lp/mm.

From my own test, I know that I did reach the upper band of resolution as an amateur, and I also know that my digital prints have higher resolution than my old negatives, actually to a degree that I can easily create 60 cm X 90 cm prints, which look sharp still even if viewed from the same distance where my old wet procedure prints just start to look unsharp.

In short, my experience doesn't seem to coincide with yours here.

Now, let's get back to the calculation, and let's say we want to achieve 40 lp/mm as system resolution, which I personally find quite reasonable, due to the findings about amateurs stated above.

In order to do so, we need a reasonably decent lens, and let's say we stop down to F/8, which will give us, with a good lens, at the Rayleigh criterion, 200 lp/mm (the lens diffraction limit).

1/sys_resolution = 1/lens_resolution + 1/sensor_resolution, in lp/mm, and under ideal circumstances.

lets fill in what we have already, and we get:
1/40 = 1/200 + 1/sensor_resolution

therefore 1/sensor_resolution = 1/40 - 1/200 = 1/50, and hence sensor resolution needs to be equal to 50 lp/mm, or 100 pixels per mm, and hence 24 * 100 * 36 * 100 = ~ 8.6 MP

Same figure, however at much smaller image dots (to be expected considering the factor 4 you used ).

The 5D II sensor is actually capable of 78 lp/mm (Nyquist limit), which is great, because in principle we should be able to get 56 lp/mm system resolution at F/8, under ideal circumstances.

I redid the calcualtions for MTF-50, and I still get a respectable 43 lp/mm for the 5D II.

In short, I don't know how and why our calculations and findings differ. I am a little puzzled.

However, this is the definition used for DoF calculations, IOW, an industry standard (12 inch print, 12 inch viewing distance, 6 lp/mm or 12 dots per mm). That is what this is really about.

This indeed makes for much smoother transitions. I also find that most people view from a comfortable distance, which is approximately 2X the diagonal of a picture - that makes for an easy viewing angle for the human eye, whilst still being able to perceive the entire image in focus. My experience anyway. IOW, a little closer for 8"X 10" prints than you seem to experience.

I have 60 cm X 90 cm (24"X 36") prints of my first 8 MP dslr. You better don't get too close indeed. However, they are still better than my older, smaller prints from colour negs, and the same sized prints from my current 21 MP camera easily beat those from my earliest dslr .

Kind regards, Wim

Good discussion. Slightly off topic......

What bugs me, more and more with time, is people posting images from different cameras, downsized to 800-1024 on the side on the web and stating, there is little difference between the cameras. True, for postage sized images. Then again, real life seems to be postage sized web images these days

harcosparky wrote in post #13438310
Well looking back is one way of finding out the minimum number of pixels needed to render a " suitable picture " depending on what is suitable for the person who wants to know.

Suitable for what is the question. Looking at a low res image on the net or printing a large detailed photo. If the former, then a 3mpix camera will do. If the later, I shoot with a 21mpix and push it's limit on very large prints.