D60's ( APS-C )sensor is too small to stop down below f/11?

Sure. It's just basic maths. 5 lp/mm on an 8x10 print (let's call it 8x12 to keep the aspect ratios the same and avoid even further confusion) amounts to a total picture resolution of (5x200) x (5x300) line pairs, or 1000 x 1500 line pairs. (On the basis that 8"=200mm).

On the basis that 3 pixels are required to record 2 lines (or 1 line pair), we need a sensor containing 3,000 x 4500 pixels (or 13.5 megapixels). If you wish to disagree with the practical evidence that 3 pixels are required for each line pair and wish to opt for a more optimistic figure of, say 2.5 pixels per line pair (as perhaps Harry does) then the sensor would have to be a 9.375 megapixel sensor.

On both counts, we clearly have a situation here where neither the D60 nor 20D (nor any sensor of any size with less than 9.375 megapixels, or 13.5mp depending on your politics) is able to produce an 8x12 print with 5 lp/mm resolution.

For that you'll need to go to the most viewed thread on the forum where people get lots of fun photographing their empty boxes of equipment purchases ("Let's see your Boxes").

Threads such as this, which require people to think a bit, are not so popular.

I see what you mean, but wouldn't then diffraction limit resolution _before_ stopping down to f/11?
Also: if this '||' is one linepair what is 5 lp?
Is it this: ||||| or this |||||||||| (first case needs 9 px , second 19 px to 'record it' (one pixel for 'line', one for 'no line'))
In the first case you need a sensor of around 2000x3000 px to record 1000x1500 lp, so I think that's what Harry uses.

Thoughts?

A line pair consists of 2 lines of different contrast. 5 lp/mm consists of 10 alternate lines of differing contrast within the width of 1mm. There should be no confusion about this.

Only the Foveon sensor can reach the Nyquist limt where one 'real' pixel (RGB) can equate to 1/2 a line pair, or one line.

I suspect that Mike is right that at f11 a lens is capable of delivering sufficient resolution to a sensor the size of the D60 or 20D to make an 8x12" print with 5 lp/mm , but the current sensors of the D60 and 20D do not have enough pixels to capture such resolution.

I'm assuming you are referring to 3 px because you need one to 'see' the black line, one the space in between, and one for the second black line?

If a lp is two black lines separated by 'white', you need 3 px for 1 lp, 9px (11?) for 5 lp. (5 lines seperated by 4 areas of white), so IMO you don't need 15px to 'see' 5 lp.
Anyhow, I'd say that I can see where the approx. 2000x3000px figure comes from and the difference by about a factor 1,5.

If by 'needing 3 px per lp' you are referring to the Bayer pattern of the sensor, things get even more complicated I think.

A line pair is one white line adjoining one black line. Two black lines separated by a white would be a line triplet.

The reason you need more than one pixel for one line is partly because you cannot rely upon perfect registration of pixels against lines and partly because of the demosaicing and interpolation processes that occur with the Bayer type systems where we have 2 green pixels for every blue and red pixel.

Ah, you say you can not rely upon perfect registration, but according to theory (and also all tests I have seen so far), a Bayer pattern and interpolation works pretty well in registering black/white differences ... provided the lens has sufficient resolving power and high acutance.

After all, black is defined as 0,0,0, and white as 256,256,256 (or whatever scaling you are using), so a pixel in a bayer pattern should read either 0 or 256 - regardless of its colour filter.

But of course, you are right in that some detail is lost - mainly because of the antialiasing filter.

The praised Foveon sensor would give you a very clear ... 3 MP image, I am afraid.

Best regards,
Andy

The imprecise registration of pixel against line is probably not the most significant reason for 3 pixels being required for 2 lines, but in a worst case scenario it would be possible for each pixel to straddle each black and white line so that each pixel would have the same value, which would result in a totally grey image with no discernible lines. On average, there's going to be some slight difference between each line and each pixel value, especially if it's a stark black and white, but if it's not, as is often the case, there's not much hope.

Those who are still skeptical about this issue of 'real' or actual resolution of Canon cameras with Bayer type sensors should check out the dpreview site, here for example http://www.dpreview.co​m/reviews/sigmasd10/pa​ge18.asp

Dpreview measures resolution in terms of lines per picture height. Note that these are lines, not line pairs. The Sigma SD10 sensor is 2268 x1512 pixels. It can manage 1550 lines of picture height (775 line pairs). This is not enough to produce an 8x12" print with 5 lp/mm, but it's a good result nevertheless. The SD10 doesn't need more than one pixel per line because it has 'real' pixels and doesn't need to interpolate values. Resolution can consequently reach, and even slightly exceed, the Nyquist limit.

The Canon 10D, after interpolation 3072x2048, has considerably more pixels than the Sigma SD10, yet it can manage only a few lines more resolution, 800 line pairs per picture height. To get 5 lp/mm on an 8x12 print, the sensor needs to deliver 1,000 lines per picture height.

However, these figures imply that not quite as many as 3 pixels are required for each line pair (2048/800=2.56) but the fact remains that no D60 image will contain sufficient resolution to produce an 8x12 print with 5 lp/mm, whatever the aperture and whatever the quality of the lens, because the D60 sensor has too few pixels.

The Pixel Density Gap

The following graphic illustrates a pixel density gap (between 188 and 291 pixels/mm) among 102 digital cameras (and backs) for which I've collected specifications.

http://home.globalcros​sing.net/~zilch0/image​s/DensityGap.jpg

I submit that this gap evidences a choice that manufacturers make when designing a sensor. That choice apparently tends to go one of two ways - toward high-density sensors or low-density sensors, with low densities going to the sensors found in the more expensive, large-sensored cameras and backs.

Ask yourself why manufacturers aren't producing full frame sensors with the higher densities common to smaller sensors.

It's simple. Vulnerability to diffraction and noise increase as pixel density increases.

Mike Davis

Mike,

interesting chart
Do you have the RAW data for that (listing of the 102 cameras) also online somewhere?
(or at least examples of some of the more popular cameras for the very lazy among us?)

EDIT: OK, found it myself (did search the thread after all):
http://home.globalcros​sing.net/~zilch0/tools​/DigiSpecs.xls
Just out of curiosity ... where did you get the sensor sizes from?
I looked up some of the cameras on dpreview.com, and there were different sensor sizes given (e.g. 4/3 system 18.00 x 13.50 mm instead of 17.3x13 in your table)

So, to sum up the difference
... below 188 pixels/mm: large-sensor (> 4/3" / 17,3x13 mm) camera (all DSLR's & digital backs, Sony R1 etc.)
... above 291 pixels/mm: small sensor (< 2/3" / 8,8x6,6 mm) camera (all P&S type cameras)

Best regards,
Andy

Hi Andy,

I always start by looking for the sensor's active area dimensions in the dpreview specifications for a given camera. If the specifications provide only the sensor type (Ex: 1/2.5"), I then go to dpreview's sensor size page to get the dimensions for that sensor:

http://www.dpreview.co​m …ystem/sensor_si​zes_01.htm

The Olympus E330 EVOLT has a "4/3" sensor, but dpreview's specifications for that camera give the dimensions of the active area of the sensor as: 17.3mm x 13.0mm. Enlargement factor, format diagonal, etc. are calculated from these dimensions.

I think the existence of the pixel density gap is interesting - it's something I've never before plotted, and I think the probability that this is just a coincidence is very low. Out of 102 sensors plotted, across a range of densities that starts at 84 and ends at 535, there's a no-man's-land right in the middle, spanning 103 units (the range from 188 to 291) where NO sensor has EVER been made (to my knowledge). This 103-unit gap is huge. It's fully 22% the height of the entire curve. (535 - 84 = 451 and 103 is 22% of 451).

It implies a conscious choice on the part of the manufacturers: Are we going to allow this camera to be vulnerable to diffraction and noise, or not? There's no sense trying to ride the fence. So in every case, the sensors are designed to go for one market or the other.

The punchline, of course, is that pixel density *is* a factor in producing a quality image. If it weren't, the manufacturers of expensive DSLR's and MF backs would already be using densities as high as those found in far less expensive digicams. Processor speed is certainly a limiting factor (it would be tough to write 100 megapixles with each frame at 8 fps), but if you're goal is to make a high quality 10 megapixel image, why are they using large sensors at densities less than 188 pixels/mm instead of smaller sensors in the DSLR's (at densities higher than 291 pixels/mm)? Answer: If you're going to capture detail (without having to shoot at your widest apertures all the time), diffraction demands a low-density sensor.

Once diffraction has caused a loss of resolution (detail), you can't get it back in Photoshop. Once diffraction has caused a loss of detail, all you can do is improve the accutance. The image can be made to appear "sharp", but the detail you could have had, with the same number of pixels on a larger sensor, simply won't be there.

Here's a good illustration of the difference between accutance and resolution:

See the "Comparison" section at: http://www.cambridgein​colour.com/tutorials/s​harpness.htm

Mike Davis

zilch0md wrote:
I always start by looking for the sensor's active area dimensions in the dpreview specifications for a given camera. If the specifications provide only the sensor type (Ex: 1/2.5"), I then go to dpreview's sensor size page to get the dimensions for that sensor:

http://www.dpreview.co​m …ystem/sensor_si​zes_01.htm

The Olympus E330 EVOLT has a "4/3" sensor, but dpreview's specifications for that camera give the dimensions of the active area of the sensor as: 17.3mm x 13.0mm. Enlargement factor, format diagonal, etc. are calculated from these dimensions.

Yes, I was assuming that you do something like this. Just wanted to verify.

Best regards,
Andy

Hi! Mike,
And I thought I'd killed off the thread!! There is a gap between the pixels of the largest P&S digicam and those of the smallest 'serious' camera, the E-330, but I doubt that this gap results from a conscious design decision involving DoF considerations. There's also a significant gap between 35mm film and the smallest MF camera, 6x4.5. I remember well when the APS film format was introduced. Who wants that, I thought. They are going the wrong direction. Give us a slightly larger format in approximately the same size body, like 36x45mm.

The fact is, if any consortium of manufacturers thought they could steal a good share of the market by creating a new format halfway between 2/3rds and 4/3rds, they'd do it. DoF doesn't come into it. Most people now know that the DoF limitations of the smaller format are not described as inadequate but too much. There is no P&S equivalent of the Canon 85/1.2. If shallow DoF is what you want, then a P&S camera is the wrong tool for the job.

Having fewer but larger pixels on the same size sensor does not get you more detail at any aperture you care to use. Just the reverse in fact. Fewer pixels on the same size sensor gets you less detail at some (most) apertures, but never more detail whatever the aperture.

When/where did I write that the gap (in pixel densities) "results from a conscious design decision involving DoF considerations"?

You've either failed to understand what I've written or you are deliberately putting a spin on what I've written.

Mike Davis