The diffraction limit is indeed something the sensor designers consider all the time. But they tend to balance aperture against photosite size for this. In other words, at f/2 the theoretical maximum resolution is achieved with photosite sizes no smaller than 2.7 microns, at f/2.8 it would be 3.8 microns, and so on. Those are the "hard" boundaries they're working against.
I like to think of noise in terms of its power spectrum -- how much noise there is at a particular spatial frequency, or scale, in the image. This is similar in spirit to an MTF chart, which plots how much contrast is transmitted at a particular spatial frequency; except here one is plotting the amount of noise at a particular spatial frequency. Read noise is largely uncorrelated from pixel to pixel (there are line or pattern noises that are correlated, but they typically make up a small component of the read noise); it is spatially white noise. This means that its statistical properties as a function of spatial frequency are entirely determined -- the noise power in a 2d power spectrum is constant for all frequencies. So pick some reference frequency well away from the Nyquist frequency, and use that as a measure of noise at a fixed reference scale.
With the WWW, finding enough suspects is not tough. It's just a simple marketing problem. ;-)
OK, I'm all ears...
Prolly, but not until we need it?
Yes, good point.
I very much agree. I think making controlled prints will be easier than trying to get several, equal monitors.
Yep, I remember the Hi-Fi years...subject matter affects the mood of the suspects and thus their choices. I found that personally I have to play SEVERAL different types of music when I'm evaluating something...and I never use something that I hate to listen to. Same way with photographic images!
I'm with you, S p i t z e r...but I'll state it differently:
Thanks. Some of the comments I made from memory, and would take back. I prefer the FZ50 version, even downsampled to 400D pixel density. Note that I did no noise reduction or noise-hiding. I simply did "as literal as possible" manual RAW conversions. A tone curve could hide some of the FZ50's red banding noise in the black areas; also, a chroma-NR'ed version of the FZ50 would probably still have a lot more resolution.
Well, if YOU ain't sure, don't expect me to help much! ;-)
Yes. I contend that Roger was honest and reasonably transparent in making his assumptions. I was too, but not everybody reads well (or completely).
It seems that the recent Panasonics, like the LX3 are better than the FZ50, with less noise, especially line noise.
I only have one...it doesn't work (but I don't think it's because of the tiny photosites.) On the other hand... ;-)
If I recall correctly, the article found a correlation between having small numbers of electrons in wells and visible artifacts.
My head is spinning already! ;-)
I think you read much more into my above sentence than I wrote into it?
I picked a small number to avoid accusations that I was unfairly biased against resolution. I didn't anticipate your stance, which is of course correct.
I have never made that connection. I think PP affects ALL the parameters that we commonly talk about: resolution, noise, and DR. When I calculated the two curves, the only variable I used was PP.
No, I got that point...I just didn't word that sentence very well. I do that late at night sometimes...and also in the morning before my first cup.
The trend of recent DSLRs to have read noise which drops rapidly in proportion to ISO going from base ISO to ISO 800 or 1600 makes the value of binning quite questionable or limited, since hardware-binning 4 photosites is basically like reading 4x as many photons, or readout at 1/4 the ISO, which has, guess what, almost the same read noise if you're talking about binning ISO 400 or 800.
Well, for the time being, most of us are looking at monitors with roughly the same PPI, so pixel-level has some semi-standardized context.
Bingo! But I think a consideration of the photographic environment as well as the print size and sensitivity setting is germaine.
I almost agree. In my diagram, I indicated that a sensor in the MIDDLE is the best all-round choice. Besides, even the low MP sensors today have enough pixels to do big prints!
My above contention is that we don't really have new cameras with low MP sensors in them any more...all modern cameras have enough pixels to do that 12 x 19 print. We have to go drag up OLD cameras to get small pixel counts (well, except for the D2H).
There is a subthread...we are going to try (again) to test these issues.
I can proudly state that only my microwave oven has a vacuum tube. ;-)
I like that suggestion, but I don't think it would eliminate this problem entirely...you have more faith in humans than I do?
Although, as Iliah Borg pointed out, most manufaturers use the standard's Recommended Exposure Index rating, which allows the manufacturer to declare the 'ISO' rating to be anything they like. This can be used to finesse the noise by setting the metering anywhere in the available raw DR, particularly if the JPEG DR is compressed. Leave lower headroom, giving more exposure, more photons = lower shot noise - hey presto.
) of the posts in that thread and others related.