More Pixel Density

Started Jul 28, 2008 | Discussions
John Sheehy Forum Pro • Posts: 21,564
More Pixel Density

Well, some people requested the 400D pixels vs the FZ50 pixels at the ISO 1600 setting on the 400D, to take advantage of the high ISO advantage with Canon CMOS, so here we go. Again, the exposure is the same, the real focal length is about the same, and the 400D is set here to ISO 1600. Actual exposure index is about ISO 2000 or a little higher for both. Just RAW interpolated, with WB based on the grey fabric under the subjects. Large images are at the FZ50's native pixel resolution, and the insets are at the 400D's native pixel resolution. 400D on the left, and FZ50 on the right:

Clearly, the 400D has an edge here in read noise even at the image level.

However, I find that no consolation, as it doesn't really help us to see what's really there any better, because of the lower resolution. The FZ50 is displaying line noises here, which, combined with the WB parameters, makes for a lot of little red streaks most visible in the dark areas. These line noises are not a necessary part of a small sensor or small pixels; they're just a sloppiness in readout, which could easily be addressed in a camera with a better electronics budget.

I have viewed the FZ50 100% crop in Lab mode and applied a 3px gaussian blur to the a and b channels, and the chromatic noise disappeared, while the luminance advantage to the FZ50 image lost no ground. NR isn't going to add resolution to the 400D versions, and sharpening the 400D upsampled version to try to make it look as sharp as the FZ50 original makes it even noisier than the FZ50 original.

IMO, the FZ50 versions, even though they have more statistical read noise, have more potential.

-- hide signature --

John

richardplondon
richardplondon Forum Pro • Posts: 10,879
Re: More Pixel Density

John Sheehy wrote:

Well, some people requested the 400D pixels vs the FZ50 pixels at the
ISO 1600 setting on the 400D, to take advantage of the high ISO
advantage with Canon CMOS, so here we go.

Mr John Sheehy, you are scaring us with your test subject here. Is that a yellow fish wearing lipstick, with cold hollow Barbie eyes? (creepy)

I am not sure I want to see it clearly...

btw, that's interesting and useful. It strengthens your case IMO to show this more typical application - both images look more like something from a real camera. The previous demonstration was a bit of a stretch for some, who suspected jiggery-pokery, which is against the Geneva Convention.

RP

 richardplondon's gear list:richardplondon's gear list
Pentax K-5 Sigma 10-20mm F4-5.6 EX DC HSM Pentax smc DA 21mm F3.2 AL Limited Pentax smc DA 70mm F2.4 AL Limited Pentax smc DA 35mm F2.8 Macro Limited +6 more
Graystar Veteran Member • Posts: 8,373
Re: More Pixel Density

Can we get the RAW files this time?

Tlon Regular Member • Posts: 271
Thank You

I've been following your recent threads re. pixel density with great interest. I'd just like to thank you for taking the time to actually test a very common assumption, sometimes facing inappropriate (to say the least) language and baffling misunderstanding.

I have two questions, though:

1. How about lens resolution? How small can the pixels become before the lens prevents further improvement in IQ?

2. The current thread shows a slight advantage for large pixels at high ISO, while the previous ones showed advantage for small pixels at low (although pushed) ISO. Can you explain when large pixels will be better and when they will be worse?

Thanks!

-- hide signature --
 Tlon's gear list:Tlon's gear list
Canon PowerShot A710 IS Sony a6300 Sony E 18-200mm F3.5-6.3 OSS Sony E 10-18mm F4 OSS Sigma 30mm F1.4 for Sony E +2 more
NIK11 Senior Member • Posts: 2,827
Thanks, also. New Pana's??

John,

Thanks for more evidence. I have followed all the threads on P density and DR issues and for me the discussion has been an education - certainly much more information than really I need to know as an ordinary punter.

What has struck me though is how you have provided samples in a scientiffic manner together with good logical explanation - very few seem to refute your findings with any conviction - and yet the widespread conventional wisdom of 'DPRE experts' still hold on to the view that new high density sensors MUST be fundamentally worse than low density sensors of old. In simplistic terms, I guess it is a bit like trying to convince people that several minors can do more manual work in a day than one olympic weight lifter - maybe true technically, but it doesn't quite fit our vision of a strong man. Divergent beliefs and facts.

There is much interest in the new Pana releases and it will be interesting to see if the heralded 10mp LX3 delivers better IQ than the 14mp FX150 - with almost identical sensor sizes. Any thoughts?

Nick

ejmartin Veteran Member • Posts: 6,274
Re: Thank You

Tlon wrote:

2. The current thread shows a slight advantage for large pixels at
high ISO, while the previous ones showed advantage for small pixels
at low (although pushed) ISO. Can you explain when large pixels will
be better and when they will be worse?

The reason for this is the ISO dependence of the cameras' electronic noise in reading out the sensor data. When referred to the number of equivalent photons' worth (equivalently, the number of electrons in the photosite) of noise it represents, the FZ50 is roughly independent of ISO. On the other hand, typical Canon DSLR's look like this plot (for the 1D3):

This means that, for a given exposure at low light levels, the DSLR will start performing better and better as the ISO is raised, while the FZ50 will remain the same. And so it can happen that the FZ50 is better at low ISO, and the DSLR better at high ISO. Why should this be, and is the DSLR's poorer performance at low ISO due to the pixels or to something else? I provided some analysis of that in a recent thread:

http://forums.dpreview.com/forums/read.asp?forum=1018&message=28749589

-- hide signature --
OP John Sheehy Forum Pro • Posts: 21,564
Re: Thanks, also. New Pana's??

NIK11 wrote:

There is much interest in the new Pana releases and it will be
interesting to see if the heralded 10mp LX3 delivers better IQ than
the 14mp FX150 - with almost identical sensor sizes. Any thoughts?

Any of a number of things can go right or wrong ... we have to wait and see.

-- hide signature --

John

OP John Sheehy Forum Pro • Posts: 21,564
Re: Thank You

Tlon wrote:

I've been following your recent threads re. pixel density with great
interest. I'd just like to thank you for taking the time to actually
test a very common assumption, sometimes facing inappropriate (to say
the least) language and baffling misunderstanding.

I have two questions, though:
1. How about lens resolution? How small can the pixels become before
the lens prevents further improvement in IQ?

That depends on the lens and how it is used. A good macro at f/2.8 could easily use 50x the pixel density of a kit lens wide open or at f/32. For some reason, people seem to be obsessed with the lowest common denominator.

If we had 200MP cameras, the cameras could be designed so that a user option forces lower output resolution or JPEG-like compression on severely over-sampled RAW captures.

2. The current thread shows a slight advantage for large pixels at
high ISO, while the previous ones showed advantage for small pixels
at low (although pushed) ISO.

If we're looking for noise itself, it is easier to find in the FZ50 image, because it's pixel read noise times the pixel pitch is higher than the 400D's, and because there is more pattern (line) noise in the FZ50. It is also easier to find detail in the FZ50 image, though, so I can't really say I'd prefer the 400D output here. The pattern noise doesn't have to be there to the extent that it is, and if I had used the G9, I think we'd see a bit less of the red streaks. I just continued with the FZ50 so I wouldn't be accused of trickery.

Can you explain when large pixels will
be better and when they will be worse?

Many CMOS sensors (complex ones; not the simple implementations used in cameras like the D2X) have lower read noise relative to absolute signal at the higher ISO. It is highest at base ISO, drops very rapidly for about two stops higher ISO, and then starts leveling off and completely levels off when the non-gain ISO are reached (usually the top one or two ISOs on DSLRs). The small sensors in compacts and bridge cameras do not have the required circuitry to do this, so noise, relative to absolute signal, is usually very close at all ISOs.
--
John

DSPographer Senior Member • Posts: 2,464
Re: More Pixel Density: Thanks

John,

This is a pretty good comparison that emphasizes the importance of shot noise even at an exposure equivalent to ISO2000 (and that the FZ50 and 400D have equivalent shot noise). Is the rug in the background in focus? I think it has interesting detail but I'm not sure that I can trust it for sharpness. If you wanted to explore the boundary of shot versus read noise contributions you could make a series of comparisons focused on the rug where the exposure is progressively reduced until neither camera records anything but black for the rug.
Thanks again. I am impressed by how well the FZ50 did.

 DSPographer's gear list:DSPographer's gear list
Canon PowerShot G7 X Canon EOS 5D Mark II Canon EF 24mm f/2.8 Canon EF 50mm f/1.8 II Canon EF 200mm f/2.8L II USM +4 more
OP John Sheehy Forum Pro • Posts: 21,564
Re: More Pixel Density: Thanks

DSPographer wrote:

John,
This is a pretty good comparison that emphasizes the importance of
shot noise even at an exposure equivalent to ISO2000 (and that the
FZ50 and 400D have equivalent shot noise).

Well, this image is a bit in the read-noise dominated range for the FZ50; if not statistically, at least visually because of the pattern noises. The pattern noises are never from shot noise; always from readout/digitization. I just realized this afternoon a slight mistake I made. The FZ50 has black-clipped RAW data, the 400D does not, and I forgot to level the playing field by black-clipping the 400D before interpolating full color, and doing WB, so the 400D pixels have a slight advantage here. This would not have happened with the G9, so I think that I am leaning towards redoing these types of shots with the G9. Canon RAW data is a lot easier to work with. And, even though the G9 has slightly more read noise and shot noise (statistically) than the FZ50 per unit of area at a given absolute exposure, it has less pattern noise. The only problem with the G9 (the reason I didn't use it in the first place for the Empire State Building crop) is that it is not easy to get the exact focal length you want with it. The FZ50 has a real, mechanical zoom ring from which I can interpolate focal lengths visually.

Is the rug in the
background in focus?

Shouldn't be very far out of focus. 22mm and f/5 for both cameras; I used the B&W striped "Schleeh" pad for focus.

I think it has interesting detail but I'm not
sure that I can trust it for sharpness. If you wanted to explore the
boundary of shot versus read noise contributions you could make a
series of comparisons focused on the rug where the exposure is
progressively reduced until neither camera records anything but black
for the rug.

Well, the shaded area under the breast of the fish approaches that, and you can see that the FZ50's line noise is too much to magnify and amplify its pixels to any great extent.

Thanks again. I am impressed by how well the FZ50 did.

Well, its the FZ50 pixels and electronics, in a "per unit of area" sense. With the same number of pixels, and FOV, the 400D would be vastly superior, using the entire sensor. I suspect that the G9 pixels might look a little bit better in the darkest areas. I was just looking today at a G9 RAW pushed from ISO 80 to about 2000 in similar light color (overcast sky), and there were no visible streaks of any color in the deep shadows.

-- hide signature --

John

Keyboard shortcuts:
FForum MMy threads