Relative Pixel Density - don't expect miracles from 40D

[isaacarus]

Or where do you find a market for lo-res small dimension prints?
I'd like to know.

most of the people ive shot for dont care about resolution.... i could shoot 1024x768? and they wouldnt be able to tell the difference really.... most of them just get resized for the web anyway...

as a sort of test camera id love to see how high an iso something like a 3-4MP 1.6 crop camera could go, i bet that would be sexay at 6400 or more, who knows you know, something like a limited run or an anniversary or something? drooooollss ;D

does anyone have any idea (mathematically) of how high an iso compared to the current cameras we could go? and the DR?

keeps drooooooling

....maybe i could build one for my 30D myself? (lol never....)

 

[Fred Briggs]

Or where do you find a market for lo-res small dimension prints?
I'd like to know.

By no stretch of the imagination can you classify the pictures you can produce from a good 8MP camera as lo-res and small dimension.

On the other hand, attempting to reproduce at large size pictures from any digital camera, if taken in adverse lighting conditions, can make it painfully obvious how rapidly image quality drops off, and noise reduction artefacts increase, with lower light.

In my opinion, the higher the pixel density of the camera, i.e the smaller the pixels, the quicker and more severe is this drop off in image quality.

In my admittedly limited professional photography career, (now ended) I have never been seriously hampered by lack of resolution, but often by low light and its adverse effects on image quality.

Of course YMMV.

BTW - I've owned and used D30, 10D, 350D, 5D, 400D, and currently 20D.

I "upgraded" from 400D to 20D as an interim camera while waiting to possibly go back to full frame with whatever replaces the 5D. Just need to convince my wife that FF is justified for non-professional use!

Fred

 

[Fred Briggs]

i don't think people expect it to match the 1dMkiii in terms of high
iso performance, but they do hope that it might do a stop better than
the 20D/30D,as the mkiii apparently does over the mkii despite going
8-> 10MP.

Yes I think sensible people hope for a small improvement over the current 20D/30D despite the resolution increase - though a whole stop seems rather optimistic.

However, I have seen a lot of posts which seem to have an expectation that simply using the improvements from the 1D3 sensor in the 40D will produce an equivalent level of high ISO performance while totally ignoring the fact that the 1D3 pixels are approx 1.6 times larger (assuming equal resolution) due to the larger sensor.

Considering the probably relatively small increases in active pixel area which the recent sensor improvements allow, the 60% increase afforded by a 1.3 sensor over a 1.6 are massive in comparison, and are bound to make a big difference in performance between the two lines of cameras.

This fact is the reality check I was trying to provide.

Fred

 

[JimH]

If all they did was use a higher resolution A/D system to digitize the same sensor signals as we've got now, then the DR would not be improved and I'd see little need for going to that higher resolution.

What would excite me would be if the sensor had the extra DR to support implementing the 14 Bit digitization at the same "bits per stop" as we have on the 20D/30D. In that case, the two extra bits would correspond to two extra stops of DR being provided.

This is supposed to be the case with the MkIII. And this is what would appeal to me.

But the only indication that the alleged 40D even possesses 14 bit resolution are the assumptions that it'll be a 10.1 MPixel sensor with a file size of 12 MBytes. So we really don't have a clue yet.

If the new sensor actually provides lower noise and somehow achieves a higher full-well capacity or they use some sneaky method to increase the high-end measuring capability, then it's possible (I'm not saying likely) that this new sensor could give us greater DR than the 30D had. And if they then digitize this data at a higher resolution to take advantage of the better range of the sensor, then that would be big news and make the 40D very attractive to me and others.

So I'm not basing my hope for 14 bit processing on the Digic III processor at all. I'm just hoping that they might give us a sensor and digitizing chain (as well as the processing to support it too) to that makes 14 bits worthwhile.

I guess my use of the word "processing" was misleading. I think of the entire analog chain as well as the A/D conversion and finally the actual number crunching all as constituting "processing" of the signal. But I didn't make that very clear in my post above.

I'm not holding my breath over any of this, though. But I will say that getting better DR and having higher resolution digitizing to support it would be fantastic.

We'll see soon enough, I guess.

--
Jim H.

 

[Steve Balcombe]

Or where do you find a market for lo-res small dimension prints?
I'd like to know.

most of the people ive shot for dont care about resolution.... i
could shoot 1024x768? and they wouldnt be able to tell the difference
really.... most of them just get resized for the web anyway...

Just as long as you understand that it's not like that for everyone!

as a sort of test camera id love to see how high an iso something
like a 3-4MP 1.6 crop camera could go, i bet that would be sexay at
6400 or more, who knows you know, something like a limited run or
an anniversary or something? drooooollss ;D

You can do a pretty fair simulation of a lower resolution sensor simply by downsampling - it's like reversing the effect of the smaller pixel pitch. It will never be quite as good as an actual low res sensor with big pixels, but it'll give you an idea.

Here's a quick demo. This is taken with the lowly 400D - a 30D would do a lot better. It's taken in a dimly-lit room at "ISO 6400" (explanation below), then downsampled to 1024 wide. This is a crop from the downsampled image:



No critiques, please, this is just a snapshot taken a few minutes ago as a demo!

"ISO 6400" means ISO 1600 and two stops of negative exposure compensation, shot RAW, pushed two stops in RAW conversion to bring back the levels. Again, not quite the same as a true in-camera ISO 6400, but near enough for our purposes.

The downsampling makes a huge difference of course - the original is pretty noisy. Here's a 1:1 crop:

 

[JimH]

I've "pushed" a few underexposed shots where I had no choice and they were annoyingly noisy. But resized for the web, you can't see any problem with them at all.

Downsizing hides a lot of the noise and it's a reasonable approximation of what you'd probably get with the larger pixel camera - not quite as good, perhaps, but a good simulation.

So if all you need are web-sized shots, you can do some fairly nasty things with what we've already got and then just downsize them. That way, we've got the high resolution camera for most uses and can just downsize our shots if we need to explore the limits of simulated high ISO shooting

--
Jim H.

 

[John Sheehy]

i don't think people expect it to match the 1dMkiii in terms of high
iso performance, but they do hope that it might do a stop better than
the 20D/30D,as the mkiii apparently does over the mkii despite going
8-> 10MP.

The 1Dmk2 is a very inefficient camera, as far as collecting photons is concerned. It leaves a lot of room to have more pixels with the same capacity, as happened with the mk3.

The lower noise at the pixel level at higher ISOs on the mk3 has nothing directly to do with photon collection, however. The improvement is in read noise, which depends more upon technology for readout than with pixel size, per se. In fact, smaller, packed pixels tend to have less read noise per unit of sensor area.

--
John

 

[John Sheehy]

You are ignoring total pixel count, total sensor size, etc. They are factors, too.

Packing smaller pixels does not ruin image quality, generally speaking.

In fact, any crop from a DSLR the size of a 10MP P&S camera's full sensor will capture what looks like garbage compared to what the P&S sensor captures, AOTBE.

--
John

 

[John Sheehy]

Yes I think sensible people hope for a small improvement over the
current 20D/30D despite the resolution increase - though a whole stop
seems rather optimistic.

However, I have seen a lot of posts which seem to have an expectation
that simply using the improvements from the 1D3 sensor in the 40D
will produce an equivalent level of high ISO performance while
totally ignoring the fact that the 1D3 pixels are approx 1.6 times
larger (assuming equal resolution) due to the larger sensor.

Considering the probably relatively small increases in active pixel
area which the recent sensor improvements allow, the 60% increase
afforded by a 1.3 sensor over a 1.6 are massive in comparison, and
are bound to make a big difference in performance between the two
lines of cameras.

The 1Dmk2 was a very inefficient design, though. It had about the same number of photons captured per pixel as the 20D/30D, at 1.7x the pixel coverage area.

The 20D/30D has just as little noise at ISO 1600 as the 1Dmk2.

It is not unreasonable to expect mk3-like noise levels at ISO 1600 on the 40D.

--
John

 

[John Sheehy]

To me, the 14 Bit system would only be really valuable if the
sensor's noise is low enough relative to it's "full well" capacity to
give us greater dynamic range than what we currently get.

I've converted 14-bit ISO 100 images from the mk3 with both the original 14 bits, and quantized first to 12 bits (and other bit depths as well). The 2 LSBs do absolutely nothing for the conversion that is visible, even pushing the deep shadows to ISO 1600 or 3200. Cameras generate far too much noise for more than 12-bits to be useful. Even 12 bits is overkill, especially at high ISOs.

--
John

 

[JimH]

Is it just marketing hype?

--
Jim H.

 

[Fred Briggs]

In fact, any crop from a DSLR the size of a 10MP P&S camera's full
sensor will capture what looks like garbage compared to what the P&S
sensor captures, AOTBE.

I'm not sure what that proves. Such a crop will capture a very low resolution portion of a much bigger and higher resolution image. How can you compare that with a full image from a P & S sensor of equal resolution using all its pixels?

Now if you magnify and compare crops containing the same number of pixels from both sensors, say at ISO 400, then I know which will appear garbage!

This is the point I am making. The quality, or to put it more technically, the accuracy of what is captured from a large pixel is going to be better than that from a small pixel, and most particularly so when photons per unit area are in short supply, pushing the signal level down closer to the noise floor.

Everything that happens from the point of capture and sampling onwards, including de-mosaicing, etc, is critically dependent on the accuracy of the original input - i.e GIGO applies (garbage in, garbage out).

If the signal level is 100 x the average noise level then you can be fairly confident in the accuracy of the capture, if it is only 2 x the average noise, then due to the random nature of noise over time you will be very uncertain of the capture.

Surely a smaller pixel will always reach the point where signal and noise are indistinguishable before a larger pixel does, and as I understand it, this and the maximum well capacity mark the two ends of the Dynamic Range available.

While there may be minor differences in noise levels between pixel sizes, surely such small differences will be swamped if one pixel has many times the light capturing area of the other.

To go back to the original post: IQ, including DR, particularly at high ISO, does seem to relate quite well to pixel size, and why else would Canon go to the expense of developing larger 1.3 and FF sensors if it made no difference?

Fred

 

[stevej615]

I've been giving this some thought, and I think there might be a firmware solution to high noise with high pixel density and high ISO. It is theoretically possible to have a 12 mp image at low ISO, and use firmware to combine the outputs from pixels to produce an effective 8 mp image at higher ISO. I don't know that Canon will do this, and I don't know how the market would react to a camera where the pixel size of an image wasn't fixed.
--
Steve

 

[SisQue]

Hi Jim, I never got to jump back into that last thread discussing this. You guys all understand the tech stuff much better than I do. But tagging on to my mention of the discussion with the Canon rep - he represented to me that the 14 bit depth increased the dynamic range. Maybe he was misinformed or trying to simplify things for me. I do know that there was a visible and distinct difference in the comparison images I viewed. Whether these differences were due to something not related to the 14 bit depth I cannot say - all I know is that is what he represented to me. Maybe this is something that cannot be accomplished in the 1.6 crop sensors - but I will be waiting to see

Cindy

--
******************************************
http://www.pbase.com/cindyd

 

[mats_b]

I see no reason why Canon could not release a newer cam with worse image quality. They do it frequently in the compact digicam segment. The 400d has softer images than the 350d at the same ISO value despite no longer havening the underrated ISO values of the 350d. Still, I see no revolt happening.

Regards Mats

 

[John Sheehy]

I've been giving this some thought, and I think there might be a
firmware solution to high noise with high pixel density and high ISO.
It is theoretically possible to have a 12 mp image at low ISO, and
use firmware to combine the outputs from pixels to produce an
effective 8 mp image at higher ISO. I don't know that Canon will do
this, and I don't know how the market would react to a camera where
the pixel size of an image wasn't fixed.

Sure, you can under-expose at a low ISO and downsample the image to reduce pixel-level noise, but you don't reduce image-level noise, and you do lose resolution, and besides, exposing properly at a higher ISO gives less read noise with Canon DSLRs, and shot noise is only determined for a given subject by the Av and Tv values, and have nothing to do with ISO settings except as ISO alters metering.

--
John

 

[John Sheehy]

Is it just marketing hype?

It could be. It may just have been that there was no point in using 12-bit, as the 14-bit ADC was no more expensive.

The only technical reason I can think of is rather indirect; perhaps the extra bit depth guarantees 2 bits more precision in RAW converters. In my experiments with altering the data in uncompressed DNGs, I found that 12-bit RAWs use 16-bit data, with the 4 MSBs unused (but you can use them re-writing the RAW data in the file). ACR apparently will use more bit depth for conversion if you simply shift the bits to more significant positions, but it has to know that the whitepoint has been raised. The extra two bits in the 14-bit RAWs may force such a bit shift, even though very little real signal is in them. Canon may have found better results in their own code with the higher level of precision forced by the two extra bits of noise.

--
John

 

[bronxbombers]

yeah, you are probably right, i think one stop is a bit too optimistic unless they stayed at 8MP. if they go to 10MP which i am sure they will do (at least) then maybe 1/2 to 2/3 of a stop, i hope.

actually more worrisome, i'm not sure i recall the 1dmkii being os much better than the 20D.... ugh. i hope it doesn't mean they will be lucky to break even if they go to 10MP.

anyway i will stop with speculation, all these finer details are rather random, even if the pics might be legit.

i don't think people expect it to match the 1dMkiii in terms of high
iso performance, but they do hope that it might do a stop better than
the 20D/30D,as the mkiii apparently does over the mkii despite going
8-> 10MP.

Yes I think sensible people hope for a small improvement over the
current 20D/30D despite the resolution increase - though a whole stop
seems rather optimistic.

However, I have seen a lot of posts which seem to have an expectation
that simply using the improvements from the 1D3 sensor in the 40D
will produce an equivalent level of high ISO performance while
totally ignoring the fact that the 1D3 pixels are approx 1.6 times
larger (assuming equal resolution) due to the larger sensor.

Considering the probably relatively small increases in active pixel
area which the recent sensor improvements allow, the 60% increase
afforded by a 1.3 sensor over a 1.6 are massive in comparison, and
are bound to make a big difference in performance between the two
lines of cameras.

This fact is the reality check I was trying to provide.

Fred

 

[bronxbombers]

they did at least keep the collectors the same size despite going to 10MP.

but yeah, i probably got way optimistic about the the 1 stop. i forgot that the 1dmkii wasn't a noise king despite having a 1.3x crop.

i don't think people expect it to match the 1dMkiii in terms of high
iso performance, but they do hope that it might do a stop better than
the 20D/30D,as the mkiii apparently does over the mkii despite going
8-> 10MP.

The 1Dmk2 is a very inefficient camera, as far as collecting photons
is concerned. It leaves a lot of room to have more pixels with the
same capacity, as happened with the mk3.

The lower noise at the pixel level at higher ISOs on the mk3 has
nothing directly to do with photon collection, however. The
improvement is in read noise, which depends more upon technology for
readout than with pixel size, per se. In fact, smaller, packed
pixels tend to have less read noise per unit of sensor area.

--
John

 

[John Sheehy]

In fact, any crop from a DSLR the size of a 10MP P&S camera's full
sensor will capture what looks like garbage compared to what the P&S
sensor captures, AOTBE.

I'm not sure what that proves.

It proves that small pixels packed into a given area give better images than a smaller number of larger pixels.

Such a crop will capture a very low
resolution portion of a much bigger and higher resolution image. How
can you compare that with a full image from a P & S sensor of equal
resolution using all its pixels?

I don't compare it, because it is completely irrelevant to the issue of pixel size, which you claim to be discussing. All the evidence to support your dismissal of small pixels is based on total sensor size, and total pixel count. Nothing at all about pixel size, which my example addresses directly.

Now if you magnify and compare crops containing the same number of
pixels from both sensors, say at ISO 400, then I know which will
appear garbage!

Yes, and ... ?

What does that prove, but that bigger sensors can capture more light within a given FOV?

This is the point I am making. The quality, or to put it more
technically, the accuracy of what is captured from a large pixel is
going to be better than that from a small pixel, and most
particularly so when photons per unit area are in short supply,
pushing the signal level down closer to the noise floor.

No. The noise floor is only relevant at 100% view; not at a consistently magnified full image. The noise floor is dictated more by read noise, which has no correlation to pixel size in the range of current technologies. The Fuji and Panasonic ZLRs with tiny sensors and tiny pixels have less read noise than most Nikon and Olympus DSLRs. Yes, lower noise floors in the tiny-sensor/pixel cameras, even though they have higher noise bellies and shoulders.

The lower pixel noise of a bigger pixel, displayed bigger, is wrong over a larger area, and it is also wrong in that it doesn't reflect the detail beneath it like more smaller pixels would; another kind of noise that generally isn't talked about.

Everything that happens from the point of capture and sampling
onwards, including de-mosaicing, etc, is critically dependent on the
accuracy of the original input - i.e GIGO applies (garbage in,
garbage out).

If the signal level is 100 x the average noise level then you can be
fairly confident in the accuracy of the capture, if it is only 2 x
the average noise, then due to the random nature of noise over time
you will be very uncertain of the capture.

Your chosen figures are a bit of an exaggeration. Tiny pixels generally have about 2x - 3x the shot noise of large pixels, and read noise can vary from lower than the big pixels, to several times higher (e.g., compared to Canon at high ISOs). However, when looked at from the perspective of noise energy per image, photon noise is unaffected by efficient division into smaller pixels, and read noise can be better than what can be achieved with larger pixels.

Surely a smaller pixel will always reach the point where signal and
noise are indistinguishable before a larger pixel does, and as I
understand it, this and the maximum well capacity mark the two ends
of the Dynamic Range available.

It doesn't matter if the noise increases, if the resolution increases accordingly. Read noise can increase along with the increase in linear resolution, with no extra image read noise, but extra resolution.

While there may be minor differences in noise levels between pixel
sizes, surely such small differences will be swamped if one pixel has
many times the light capturing area of the other.

To go back to the original post: IQ, including DR, particularly at
high ISO, does seem to relate quite well to pixel size, and why else
would Canon go to the expense of developing larger 1.3 and FF sensors
if it made no difference?

Your reasoning seems to be stuck in a rut; you're talking about sensor size, to evaluate pixel size. The two are not related by necessity; only by coincidence.

--
John