Apparently size really does matter or does it . . .

[edlaurpic]

Advertisement (gone after registration)

Apparently size really does matter, in more ways than one, or does it? . . .

 

Most of us acknowledge the importance of size and weight in a camera/lens combination, and we also realize that when it comes to a digital photograph, with some limitations, larger sensors often produce better pictures, but not always. But what of the importance (or lack of it) of pixel size (note, I said PIXEL SIZE, NOT SENSOR SIZE)?

 

Whenever I need a compact and light camera, I take an M8, but for tele and macro I take a DSLR (R9/DMR or 5D). However, if resolution and image quality are of paramount importance, I take a Hasselblad with a digital back (a 16 mp CFV or the 22MP 132C). Now as I am considering getting a 39MP Hass back and am still expanding my M8 kit, the question of pixel size has come up. Specifically, I have come across several provocative postings on the web about pixel size. Before I go on, if these posts are right, pixel size sounds like it is just as important for the M8 and DMR (and future R10), and I am wondering just how large the pixels are in the M8 and DMR and whether what is being said about MF backs is equally true of our Leicas (seems that it would be to some extent, but I don’t know how much). Let me explain.

 

What I came across were two postings on diglloyd.com, which, together, assert that the image quality of the 16mp Hasselblad 503 CWD with pixels that are 8.95 microns has superior image quality—on a pixel by pixel basis—than the 39MP Hasselblad H3D-39, which costs three times more, but which has pixels that are only 6.8 micros square (whatever that means). Specifically he asserts that even though the H3D-39 has higher resolution, it lacks the dynamic range and lower noise that characterize 503 CWD.

 

The url for the two posts are here: diglloyd: An Hour with the Hasselblad 503 CWD

And

diglloyd: An Hour with the Hasselblad H3D-39

 

Totally confused, given that all of the pros I know who have been using the 22 meg MF backs are switching to larger backs now (31 or 39), I called a pro for whom I have the highest respect and who I know owns the 503 CWD, the H3D-39 as well as the M8 and DMR, and several Canon DSLR’s and I put it to him— Does pixel size matter as much as diglloyd.com says it does?

 

His answer, however sincere, was less than fully responsive. What he said was essentially this: maybe on a pixel by pixel basis the image quality of a large pixel in the 503CWD is better than that of a smaller pixel in an H3D, but the increased resolution afforded by the 39 MP back is so extraordinary and he would never part with it. Yet still loves his 16MP CWD, but this might just be because of the sentimental value of continuing to use a V system camera. I should also add that he still uses the DMR and now the M8 for many events he shoots.

 

So I am also wondering whether the terrific image quality we get in files from the M8 and DMR has something to do, not only with the lack of an AA filter, but also with the size of their pixels. Now I am really feeling ignorant, because I don’t know how large the pixels of the M8 and DMR are as compared to, for example, the 5D or the new DSLR king, the Canon 1D Mark III. Does anybody know? Does it matter?

 

Before somebody says it, I realize that sensor size and image size and the presence or lack of an AA filter are all important, but so is the algorithm in the camera and a lot of other factors, but that isn’t a good answer to this issue. Put another way, other things being equal, how important is pixel size?

 

This is important to me because, much as I would love to be able to blow up a little piece of a file that a 39MP back would produce, I also love the images that the CWD produces, and I am thinking, why not get the 22MP CF instead for my Hasselblad.

 

Back to Leica, it’s not as though we have choice of sensors for our Leica digital cameras, but I am still curious about this whole area and would love to know the following:

 

1. Do our M8s, relatively speaking, have smaller or larger pixels compared to, e.g. the Canons?

2. Same question for the DMR

3. Ignoring price, if you had a hasselblad and could have either the 22CF or the 39CF, which would you choose?

4. Put another way, is Diglloyd just wrong about pixel size?

 

 

Finally, if you are really interested in this topic, check this out

 

isl.stanford.edu/~abbas/group/papers_and_pub/pixelsize.pdf

 

… which is a pdf from the Stanford Information Systems Laboratory that says, in so many words, that pixel size really does matter in the following way: – small pixel size and large pixel count make for better spatial resolution and MTF, while large pixel size means better dynamic range and SN ratio. Lastly the Stanford guys also said they believed that a socalled “optimal pixel size” may exist and they seem to be saying that that size is around 6.5 µm with fill factor of 30%.” If they are right, then maybe the 6.5 micron pixel size 39 CF still delivers terrific image quality even though its pixels are smaller than the 8.95 micron pixels of the CWD.

[lars_bergquist]

Ed, those Stanford blokes are right, I presume. Large pixel cells can collect more photons and transmogrify them into electrons, thus the video signal is stronger and has more dynamic range and needs less amplification. 160 is the true 'native' light sensitivity of the M8 sensor. Many compact digital cameras with their smaller pixel cells need lots of amplification in order just to reach ISO 100. So their dynamic range is less, amplification even at low ISO means more noise, and ISO does of necessity stop at 400 or so.

 

Even so, there is a limit to the dynamic range we can use (though we are not there yet) and beyond that point more resolution is better (but there is of course a limit to the resolution we can use too – and for some applications we are already there).

 

The old man from the Age of Pencil Sharpeners

[guywalder]

Ed,

I think the answer is: it all depends.....

The debate has raged between Canon and Nikon devotees ever since the 'full format' sensor appeared.

Certainly a larger pixel will help achieve a better S/N ratio as it can collect more photons. And at some point a pixel will presumably become so big that it hurts resolution.

However there has been a steady trend in the '35mm' digital world towards smaller pixels (i.e. higher resolution), but at the same time an improvement in high ISO noise performance. In other words, pixel size is not the only factor.

Expressed another way, all things being equal, bigger pixels will result in less noise. However all things are never equal so it is pretty much a waste of time to get hung up on pixel size!

If you really want big pixels, go get a Nikon D2h, just dont expect its other IQ qualities to compare to your M8....

Guy

ps, take a look at DP Review, there are specs which give you sensor size and number of pixels, you can work out the pixel pitch, which 'might' be indicative of pixel size...

[carstenw]

The pixel pitch is indicative, but not a guarantee. It depends on the diameter and distance of the microlens too. It would be a lot of work to tabulate real pixel sizes though, so estimating with pixel pitch ought to be close for most purposes. The 5D has some of the largest pixels in current cameras. The M8 pixels are smaller, closer to the 1Ds2. I don't recall the numbers any more, but I looked at all of this when I switched, and concluded that they were close enough for my purposes.

[guywalder]

Carsten,

I'd go further and say that pixel pitch is interesting but not indicative.

The photons which get absorbed, reflected, focussed by the micro lens array, the number of photons which get filtered out by the Bayer filter array, the fill factor of the pixel itself, and of course the processing of the subsequent signal seem to drown out the simple pixel pitch.

At the end of the day, compare the pictures, isnt that what counts?

Guy

[andit]

Hi Ed,

 

This is one of those interesting questions. Just to give you a couple of figures on the actual pixel size of various camera's. The Canon 30D has 6.4 microns, the 5D (full frame sensor) has 8.2, DMR and M8 have 6.8, Nikon D80 has 6.1, Leica Digilux 3 (or Panasonic L1) has 5.5 microns.

 

Once you look at the compact digital cameras, the pixel size becomes tiny in comparison to SLR type camera's. The Canon Ixus 800 has 3.2 microns, Leica C-Lux 1 has 2.6 microns, the Panasonic FZ50 has 3.1.

 

Even though some of the compact's have more pixels than the SLR type's, the images that you capture on the SLR type's are far superior. The smaller the actual "pixel real estate" (thats the jargon for the actual pixel size), the less light sensitive it becomes and thus produces more noise.

 

Please don't get me wrong. I use a C-Lux 1 as a daily travel camera and have taken some amazing shots with it. For real fine details and huge prints I use my M8. Sorry have not used MF for a real long time, so I cannot say anything about the Blad equipment.

 

Hope that this has helped you a bit.

 

Andreas

[edlaurpic]

Advertisement (gone after registration)

I had a hunch I could count on some of you to have some insights, and what you said is really helpful. But what do any of you make of the Stanford researchers opinion that there is an "optimum pixel size" which they believe is 6.5 microns? If they are right then perhaps it's not just a coincidence that the size of both the M8's and those of the new Hasselblad 39 backs is 6.8.

 

Maybe, as one of you just said, beyond 6.8 microns there are diminishing returns re dynamic range and noise reduction while 6.8 is small enough to allow lots of pixels and therefore better resolution and MTF. What do you think?

[carstenw]

Perhaps diminishing returns, but still improvements. The 5D definitely has advantages over the M8 in tonal values and high ISO. Its shortcomings are due to other design decisions, like CMOS versus CCD and the anti-aliasing filter. There is still a place for larger pixels. Not everyone wants the best bang for the buck, just the best bang.

[Pham Minh Son]

If you are looking for color accuracy for skin tone, you will find that the color is best achieved with 16, 18 and 22 megapixels. The 39 megapixels has the tendancy to show color shift. In addition, of the 16, 18, and 22 megapixels which are all 9 micron size sensors, the only 6.8 micron sensor that I would trust for skin rendition would be the Phase One P30. I do not see fashion photographers shooting with the 39 megapixels; landscape and architectural photographers tends to love the 39 megapixels backs.

 

-Son

[ho_co]

Ed--

You've probably already looked at more data on the topic than you had any interest in, but I'll give a couple additional pointers here.

 

First, remember there's a difference between pixel _size_ and pixel _pitch_. You can take pixels of a given diameter and place them on the substrate closer or further. The distance between centers is the pitch. This brings up another point: CMOS sensors place the leads to read the pixels alongside the pixels, taking up a certain amount of space; CCD sensors take the data sequentially from the end of a row of pixels and can place their pixels closer together (finer pitch with possibly same size pixels).

 

Second, check out "Form Follows Format" in LFI 3/2006, pp 40-47. It covers a lot of this ground in passing, particularly in regard to special aspects raised by the pixels on very small sensors, e.g. those on D-Lux 2 and Digilux 2.

 

Third, there is a lot of good information in the contentious posts and counter-posts at Luminous landscape. In this article, both writers agree that it is physically impossible for most of the new generation multi-megapixel digicams ever to achieve the resolution their pixel count implies.

 

This digital world is opening a lot of unexpected twists and turns in our understanding of optics!

 

--HC

[Likaleica]

Ed,

 

Here's a great article that I think will help you tremendously:

 

http://www.luminous-landscape.com/essays/back-testing.shtml

[lars_bergquist]

The truth is (to return home from MF-land) that we have reached the point where the technology is becoming mature and the order of the day is trade-offs – you gain some and you lose some. There are rumours of new types of colour filtering which may be more economical of light. But drastic improvements are not coming. Depending on what you mean by drastic, of course.

—What we have not considered yet is the human perceptual apparatus, a.k.a. vision. I would dare the quasi-biblical proposal that people are not made for photography, but photographs are made for people. So, what can we take in? It goes without saying that we can print our files to dizzying magnifications. But what can we take in?

—There is an old rule of thumb that at normal reading/holding distance, we can resolve two lines 1/10th of a millimeter apart. If the resolution is .1mm or more, we find the mage sharp. So, how large is the largest print we can take in at one look at that distance? A more general answer is of course given in terms of degrees, but I think a practical answer, at a distance of 30cm (close to the close focusing limit of human eyes) is something like A4. If we see a picture at a greater angle than that, and want to see the whole picture, we must back off in order to reduce the angle that the picture subtends.

—We can of course size the print to two by three feet, or meters, or kilometers in theory, and then go at it with magnifiers, microscopes, hammer and tongs. But is this a realistic use of a photograph? Yes if it depicts a scientific specimen maybe, and we want to see a maximum of detail, not a picture. But this is not how I normally use a print.

—So, show me a sharp A4 and I see a picture with enough resolution. Then we can talk contrast, dynamic range, noise, colour rendition and all that. Everything else is a technical exercise for its own sake.

 

The old man from the Age of the Focomat

[adan]

I think the "nut" of this thread is expressed above in the mention of "optimum" pixel size - i.e. some size where there is a Min/Max in the curves defining resolution and noise as functions of size.

 

But what is "optimum" depends on the situation, so I don't know that 6.5 microns IS that optimum without knowing what situation the researchers were using as a model. A Min/Max can end up as sort of an average - the "best noise for the resolution" or vice versa.

 

And the average is not the best for everyone, as the comment about fashion vs. architecture photgraphers demonstrates. For some, the extremes (of noise preformance or resolution performance) are more useful.

 

Personally and practically, I find that a sensor in the 10-13 Mpixel range is sufficient to handle practically everything a lens can throw at it. Whereas I find most sensors smaller than classic 35mm full-frame, in that pixel-count range, are noisier than I would really like for low-light work (including the M8).

 

So MY optimum - in a full-frame sensor - would be for bigger pixels rather than more pixels. I.E. - I'd prefer the R10 or M9, if they are in fact FF cameras, to be in the "family" of the Nikon D3 and Canon 5D (extraordinary low-noise, about 12-13 Mpixels), rather than trying to squeeze in more smaller pixels like the Canon 1DsIII.

[rosuna]

6,8 microns of pixel spacing leads to 18,6 millions of pixels on a 24x26mm sensor. The M8 and DMR have 6,8 microns pixels, this is, a sampling frequency of 75 lp/mm. In theory, it is possible to have 18,6 millions of total pixels (not effective pixels) and very low noise on a 35mm full size sensor.

 

Canon will do this with 6,4 microns (and a CMOS sensor instead of CCD, and this means pixel spacing isn't pixel size) in the 1Ds Mark III (22MP), and the 1D Mark III (6,8 microns, 10MP, crop x1.3, like the M8 and DMR) performs incredibly well at high ISOs. Then, 22 MP is not too much in terms of density. It guarantees 40 lp/mm cleanly resolved, the typical target for 35mm format.

 

You need a minium pixel density because you need a minimum sampling frequency in order to resolve without moiree a particular level of detail. Nyquist frequency is a half of the sampling frequency (aprox), and if you want 40 lp/mm clearly resolved you will need a sampling frequency of 80 lp/mm or a bit more (Bayer interpolation, etc.) on the sensor.

 

Leica don't use AA filters, and therefore they shouldn't reduce the density too much. My point is that keeping the density of the M8/DMR you can have 18MP in a full frame sensor, increasing it just a bit (6.8 microns vs 6.4) you have 22MP. The 18-22 range is the optimal one for final resolution on 35mm format (lens + camera), and it is very good for noise (at least for Canon's CMOS).

 

In my opinion, Leica should offer 18Mp (aprox) in a future R10 FF camera and work on better noise reduction algoritms and dedicated components on the logic board.

[sandymc]

Just to clarify what the Stanford folks were saying: They were saying that for any particular technology, there is an optimal pixel size, not that there is an optimal pixel size full stop. And by technology they mean both CMOS vs CCD vs whatever alse, and the feature size. The 6.5 mention above is specific to .35 micron CMOS technology. (For those confused by this, the .35 is roughly (very roughly!) the smallest feature on the sensor). If you built the sensor with .18 technology, then the optimal pixel size goes down, to approx 4.5. CCD sensors, as used by the M8, etc would have a different optimal size.

 

And BTW, this is a pretty old (by the standards of the semiconductor industry anyway) paper - January 2000

 

Sandy

[rosuna]

Can you bring a link to that paper? It is very interesting.

[sandymc]

Can you bring a link to that paper? It is very interesting.

 

It was in the first post, but not quite formatted right:

 

http://isl.stanford.edu/~abbas/group/papers_and_pub/pixelsize.pdf

 

Sandy

[rosuna]

Thanks

[Guest guy_mancuso]

 

Leica don't use AA filters, and therefore they shouldn't reduce the density too much. My point is that keeping the density of the M8/DMR you can have 18MP in a full frame sensor, increasing it just a bit (6.8 microns vs 6.4) you have 22MP. The 18-22 range is the optimal one for final resolution on 35mm format (lens + camera), and it is very good for noise (at least for Canon's CMOS).

 

In my opinion, Leica should offer 18Mp (aprox) in a future R10 FF camera and work on better noise reduction algoritms and dedicated components on the logic board.

 

That is something i am hoping they are doing is looking squarely at 18mpx or 22mpx for the R10 and maybe even a M9. Also hitting the ISO1250 mark as well as it is on the M8 at 640 for noise, i'm basically looking for that one extra stop on the clean noise side. Than i think leica will have a good balance of mpx vs noise without completely reinventing the wheel which takes too much time. I would rather see this sooner than later. Also we don't want to give up any DR on it either and if anything go bigger but if keeping the pixel size at this level it is now or better to improve DR than i am all for it also as long as it does not negatively affect something else. i would rather the R10 and M9 to act more MF than anything else.

[Advertisement]

Hallo,

Look here: Leica M

[rosuna]

Dynamic range and tonal resolution are very, very important. I hope Leica can find more powerful processors for handling true 16 bit files, or for better tonal compression. The idea of no-linear tonal compression is good, but the 8-bit compression of the M8 seems to me excessive. The M8 and DRM are very good cameras, so they don't need to invent the wheel but only improving. Newer technologies will help here.