The 10mp sensors are lifeless

[chuxter]

gheil wrote:
Japan Inc is playing us all for fools.

No. They know we are fools. Look at the products we buy from them...

By my count, 99% of the camera buying public buys into the "more MP is better" ploy. Every time they jump the MP up, the 1% crowd lambasts them. But everybody else races to the store (figuratively) to buy 'em.

If you don't like this statement, you are probably part of the 1%...

--
Charlie Davis
Nikon 5700 & Sony R1
CATS #25
PAS Scribe @ http://www.here-ugo.com/PAS_List.htm
HomePage: http://www.1derful.info
'I brake for pixels...'

 

[Kenstrain]

More pixels can improve highlight definition, where noise is not an issue (lens quality allowing). Usually this comes at the cost of light collection efficiency (which is in the first place the main advantage of digital capture IMO). This is due to the dead-space among the pixels, as without the dead-space it would be possible to resample the image to a lower pixel count with very little loss (and in particular to recover the noise statistics of a lower pixel-count sensor). So without dead-space more pixels with sensible processing would be harmless, but that is not the real world.

Sometimes, however, a new chip/microlens layout comes along, increasing the light collection efficiency a little, and reducing the penalty from the dead-space. When this is done having more pixels can be better. (A good example is the jump from the EOS 300D to the EOS400D - I have both and have convinced myself that the advantages of 10MP outweigh the disadvantages.)

As is often the case in this marketplace, the really important quality measures are hiden from the customer - sensor size and light collection efficiency should be the main parameters on which manufacturers compete.

Except, perhaps, by implication from noise statistics in RAW images (rarely available), or just maybe by considerations of native ISO, it seems impossible to discover the light collection efficiency of competing sensors. There should just be a clear statement: e.g. this camera has an APS-C sensor with xx% quantum efficicency on the YYY standard (I do not think there is a real standard for this.)
Pixel count can then be considered at the next level.

I realise this ignores the significance of noise reduction algorithms when in-camera jpgs are used, and is obviously not going to happen in the real world anyway (I can dream).

Perhaps the best comparison would be a standardised "information content" measure. Such a measure could probably be constructed using a test chart with perhaps a pseudo-random pattern. Analysis of the resulting image could reveal the preservation of signal data over noise.

This would show whether more pixels were adding or subtracting from the image content.

Unfortunately it could be months of work to construct such a test and make it reliable (I'm not tempted).

Your 99% may be correct, but there are also the 0.01% (?) who dream about an informed market!

Ken

By my count, 99% of the camera buying public buys into the "more MP
is better" ploy. Every time they jump the MP up, the 1% crowd
lambasts them. But everybody else races to the store (figuratively)
to buy 'em.

If you don't like this statement, you are probably part of the 1%...

 

[John Sheehy]

Except, perhaps, by implication from noise statistics in RAW images
(rarely available),

This is pretty easy to collect. All you have to do is take very out-of-focus shots of an evenly lit color-checker or grey step card, separate the RAW color channels, and find the mean and sigma values for the center of each rectangle (avoiding the blurred edges). With my Canon 20D, these values are extremely close (within a few percent) of what you'd calculate by the formula:

Nt = (photons + (Nr)^2)^0.5, where Nt is total noise and Nr is readout noise, all in electrons. Then convert back to ADUs if needed.

or just maybe by considerations of native ISO,
it seems impossible to discover the light collection efficiency of
competing sensors. There should just be a clear statement: e.g.
this camera has an APS-C sensor with xx% quantum efficicency on the
YYY standard (I do not think there is a real standard for this.)

With a consistent light source like a panel strobe, you could use t-mount tripod-mounted lenses on all DSLR cameras and make them totally passive to light with no control over exposure, and see where the light levels register in ADUs, and how much noise they have, etc.

--
John

 

[tbcass]

--
Tom

http://www.flickr.com/photos/25301400@N00/

 

[tbcass]

--
Tom

http://www.flickr.com/photos/25301400@N00/

 

[Jay Turberville]

The answer is 6, 12, 8. Two of you got it right. I'm a bit
disappointed, though, that the initiator of the thread ducked the
question instead, and simply claimed "years of experience" instead.

I'm sure that there are reasons that the initiator could give for you test being invalid. I think that for some people, "blind" testing is automatically invalid.

In fairness though, maybe he just hasn't been online lately?

--
Jay Turberville
http://www.jayandwanda.com

 

[TrapperJohn]

We, as in the general buying public. We are the ones who are demanding more MP, more ISO, instead of more vibrant images. The buyers are bcoming less like Ansel Adams and more like Bill Gates.

The camera makers are simply responding to the market.

 

[John Doyle]

The buyers are bcoming less like Ansel Adams and more like Bill Gates.

What has Bill Gates got to do with it?

 

[Kenstrain]

Except, perhaps, by implication from noise statistics in RAW images
(rarely available),

This is pretty easy to collect.

Yes, unfortunately my sentence was ambiguous - I meant that raw images are not always available, when you have one the measurement is, as you explain, usually quite simple. What I should really have said is that you need a raw file to ensure a meaningful result.

Ken

 

[Robert 55]

The buyers are bcoming less like Ansel Adams and more like Bill Gates.

What has Bill Gates got to do with it?

 

[Petteri Sulonen]

No, he replied to my message but declined to take the "test."

And yeah, I know it's invalid in any number of ways -- but I think that if there is some really significant difference between the sensors, it shouldn't be that difficult to see it.

Petteri
--
[ http://www.prime-junta.net/ ]
[ http://p-on-p.blogspot.com/ ]

 

[John Doyle]

No, he replied to my message but declined to take the "test."

And yeah, I know it's invalid in any number of ways -- but I think
that if there is some really significant difference between the
sensors, it shouldn't be that difficult to see it.

Well I count 4 people who guessed at your test so the sample size is a bit small. However everyone agreed that the last image was from the 8MP sensor. The votes were split 50% on the 12MP sensor. If the 12MP camera was the 5D, the fact that it has a huge sensor means that the pixel denisty is about the same as the density on the 6MP, which would explain the confusion, or no?

Even with the small sample the votes were remarkably consistent. Maybe it isn't that difficult to see after all.

 

[Ted1212]

I think you are right Aarif,

I've been looking around the net for the best lenses that I could find for portrait photography, not technically, but based on the images. I'm looking for a lens that creates beautiful images, with bokeh and light that makes the picture look like a work of art.

I found three lenses that do the job best in the way I'd like to see it done (there could be more, that I haven't found yet) these are:
Canon 85L, 135L and Pentax 77mm Limited.

The best Canon images with these lenses generally came from full frame cameras while the Pentax images IMO exceeded even those for what I was looking for, I was surprised to find that they were from a 6mp camera.

I was about to buy the Pentax K10D and viewed the new images posted on the Pentax website - I was disappointed when I saw them, yes the same lifeless kind of images I'd seen from the other 10mp cameras stared back at me. My wife had the same response too.

Now it may be that there are little or no differences between 'standard' pictures size for size in 6 to 12mp cameras viewed at the same size, but I know what I want to get out of my camera - art, and from what I've seen it looks like the 10mp cameras will take away an important 'edge' for what I want to do.

 

[chuxter]

More pixels can improve highlight definition, where noise is not an
issue (lens quality allowing). Usually this comes at the cost of
light collection efficiency (which is in the first place the main
advantage of digital capture IMO). This is due to the dead-space
among the pixels, as without the dead-space it would be possible to
resample the image to a lower pixel count with very little loss
(and in particular to recover the noise statistics of a lower
pixel-count sensor). So without dead-space more pixels with
sensible processing would be harmless, but that is not the real
world.

I don't have a clue what you are talking about: "...without the dead-space it would be possible to resample the image...with very little loss..." You may have a great point but you didn't explain it well enough that I can tell...sorry.

Sometimes, however, a new chip/microlens layout comes along,
increasing the light collection efficiency a little, and reducing
the penalty from the dead-space. When this is done having more
pixels can be better. (A good example is the jump from the EOS
300D to the EOS400D - I have both and have convinced myself that
the advantages of 10MP outweigh the disadvantages.)

I think the only negative to more pixels is that it tends to reduce the size of the photosite. I think my R1 with 10 MP is OK, but a 1/1.8" sensor with 10 MP is terrible. The difference is not the MP...it's the photosite area! Micro lenses don't "fix" that problem.

As is often the case in this marketplace, the really important
quality measures are hiden from the customer - sensor size and
light collection efficiency should be the main parameters on which
manufacturers compete.

If by "sensor size" you mean physical size as in mm^2, then I agree. if by "light collection efficiency" you mean the physical size of the photosites should be large, then I agree again.

Except, perhaps, by implication from noise statistics in RAW images
(rarely available), or just maybe by considerations of native ISO,
it seems impossible to discover the light collection efficiency of
competing sensors. There should just be a clear statement: e.g.
this camera has an APS-C sensor with xx% quantum efficicency on the
YYY standard (I do not think there is a real standard for this.)
Pixel count can then be considered at the next level.

I agree. There is no standard and if there was, the whole concept of "quantum efficiency" would totally miss those 99% guys! They can barely understand "area"... ;-)

I realise this ignores the significance of noise reduction
algorithms when in-camera jpgs are used, and is obviously not going
to happen in the real world anyway (I can dream).

Perhaps the best comparison would be a standardised "information
content" measure. Such a measure could probably be constructed
using a test chart with perhaps a pseudo-random pattern. Analysis
of the resulting image could reveal the preservation of signal data
over noise.
This would show whether more pixels were adding or subtracting from
the image content.

I have noticed that the size of a JPEG file sorta tells us what the information content (detail) is. Don't laugh. I often take several pix...I do not bracket exposure or WB...just take several pix. Like if I'm taking a pic of an insect or flower, I can look to see if the insect moved. But if the insect is still in all the shots, how do I pick one to PP? I look at the file size and pick the biggest one, because it has more detail than the others. Note that this only works where the JPEG parameters are constant, like all the files came from one camera. If you mix in different cameras or files written by PS, these JPEG parameters WILL be different and will produce vastly different file sizes that tell us nothing about the detail preserved in the pic.

Unfortunately it could be months of work to construct such a test
and make it reliable (I'm not tempted).

Your 99% may be correct, but there are also the 0.01% (?) who dream
about an informed market!

I'm a realist and a pessimist. ;-)

By my count, 99% of the camera buying public buys into the "more MP
is better" ploy. Every time they jump the MP up, the 1% crowd
lambasts them. But everybody else races to the store (figuratively)
to buy 'em.

If you don't like this statement, you are probably part of the 1%...

--
Charlie Davis
Nikon 5700 & Sony R1
CATS #25
PAS Scribe @ http://www.here-ugo.com/PAS_List.htm
HomePage: http://www.1derful.info
'I brake for pixels...'

 

[celltech161]

I don’t know if I’m alone but from what I’ve seen in all the new
dslrs the images lack life and are flat.

That's what I was thinking when I started seeing your A100 work. It just didn't have that extra zip that the 7D shots had.

I was rather surprised to see your comment about it in the archive, but I agree fully with you.

I see you have a Canon 5D now? How did that come about? Welcome to the club!!

 

[digital_ray_of_light]

Perhaps the Emperor should consider wearing three-layers of clothes instead ;-)

Sony has no superchip, they cannot create the dynamic range of big
sensels with much smaller ones! It is Sony that made all these
(D80, A100 and K10D) 10MP sensors. The camera makers are being lead
down the Megalpixel niny garden path by the Emperor, Sony. Canon
with its 400D is just playing along with the game "see we can make
10MP too, hope you like it". Japan Inc is playing us all for fools.
Someone has to say it: "The Emperor has no clothes".
--
http://public.xdi.org/=greg.heil

--
Comprehensive Photokina 2006 speculation: http://photographyetc.livejournal.com

 

[Kenstrain]

So without dead-space more pixels with
sensible processing would be harmless, but that is not the real
world.

I don't have a clue what you are talking about: "...without the
dead-space it would be possible to resample the image...with very
little loss..." You may have a great point but you didn't explain
it well enough that I can tell...sorry.

I'll try harder to be clear then! I meant that, if the whole surface of the sensor is sensitive (i.e. no gaps among the pixels) it does not matter if you, for example, cut all the pixels in half to double the number - the same light will be collected and you can add the signal from the two halves in software to recover the same image quality.

(Of course this needs care to keep the electronics good enough, but it can be done within reason, and this is only a hypothetical example.)

The point is that it is not that small pixels are bad per se, but that if the pixels are too small the dead-space among them becomes too significant and the overall light collecton ability of the sensor degrades.

I.e. the number of useful mm^2 . In every other way small pixels are better for the sophisticated user.

I think the only negative to more pixels is that it tends to reduce
the size of the photosite. I think my R1 with 10 MP is OK, but a
1/1.8" sensor with 10 MP is terrible. The difference is not the
MP...it's the photosite area! Micro lenses don't "fix" that problem.

But microlenses that caught all the light plus intelligent binning of the pixels to reduce the resolution and noise in an optimum way could fix the problem in principle. OTOH I agree 100% that with current designs the 1/1.8" sensor with more than a few MP is not very good (never mind the now popular and totally hopeless 1/2.5" ones).

If you compare 2/3" sensors (Sony 8MP) with APS-C (Sony 10MP), however, the performance nearly exactly scales with area (i.e the noise is equivalent at about 3 or 4 times the ISO on the larger sensor, at least roughly, and ignoring pattern noise seen in some cameras).

If by "sensor size" you mean physical size as in mm^2, then I
agree. if by "light collection efficiency" you mean the physical
size of the photosites should be large, then I agree again.

By light collection efficiency I meant the number of photo-electrons measured compared to the number of photons incident on the whole sensor (i.e total quantum efficiency, with no allowance for dead-space).

I agree. There is no standard and if there was, the whole concept
of "quantum efficiency" would totally miss those 99% guys! They can
barely understand "area"... ;-)

OTOH if Phil had a standard test that rated cameras on a scale of 0-10 for "information capture" it would probably be easier to make use of than the noise measurements, where the effects of NR on sharpness are hard to see. (I'm always very suspicious of those big dips in noise at 800 ISO etc. - I trust the measurements but not the in camera processing.)

I have noticed that the size of a JPEG file sorta tells us what the
information content (detail) is. Don't laugh.

I'd be laughing at myself too - I used file size tool, and it is reliable for fixed camera and ISO. As you know higher noise means larger JPEG files too, but then NR means smaller ones whether it removes noise or image detail, so something else is needed to compare cameras. Perhaps it exists already, but I do not remember seeing it.

Ken

 

[GordonBGood]

I'd be laughing at myself too - I used file size tool, and it is
reliable for fixed camera and ISO. As you know higher noise means
larger JPEG files too, but then NR means smaller ones whether it
removes noise or image detail, so something else is needed to
compare cameras. Perhaps it exists already, but I do not remember
seeing it.

Ken and chuxter, I've been working at a mathematical tool that would determine how much NR/loss of detail there is in a given image. I think that in a similar way that noise measurements are made on a flat textured (hopefully zero natural variation) test patch, a determination could be made on a finely detailed low contrast test patch. I have run images of such patches through the Discreet Cosine Transformation algorithm (DCT, as is used in JPEG lossy compression to apply different quanitizations to different "frequencies" of detail) . Given that the components that come out of the DCT algorithm have magnitudes that are proportional to frequency, it seems that comparing the magnitudes of the higher components to the middle "frequency" components does co-relate to the amount of NR/loss of detail.

I have found this is an adequate tool for determining how much NR/loss of detail increases with increasing ISO sensitivity in the same camera, but unfortunately the numbers obtained aren't absolute in comparing different cameras due to the variation in Bayer interpolation algorithms in how much averaging they do across wider ranges of pixels, although I suppose that could be considered to be a form of NR with potential loss of fine low contrast detail (ie. Fuji processing on "angled" sensors does quite a lot of this). I suppose that I should determine the ratio of frequencies for a very simple, maximum detail, Bayer interpolation algorithm and use that as a "standard" for rating all other camera models.

I'm sure that Phil Askey on this site would be interested in using such a tool in his reviews, as it appears that NR is becoming more an more of a factor on the overall IQ rating and the "standard" standard deviation noise measurements no longer really apply after the test patches are "smudged" by such.

Regards, GordonBGood

 

[chuxter]

Ken and chuxter, I've been working at a mathematical tool that
would determine how much NR/loss of detail there is in a given
image.

Wonderful! I'd love to assist by doing Beta Testing...

--
Charlie Davis
Nikon 5700 & Sony R1
CATS #25
PAS Scribe @ http://www.here-ugo.com/PAS_List.htm
HomePage: http://www.1derful.info
'I brake for pixels...'

 

[aarif]

I don’t know if I’m alone but from what I’ve seen in all the new
dslrs the images lack life and are flat.

That's what I was thinking when I started seeing your A100 work.
It just didn't have that extra zip that the 7D shots had.

The 5D I got in an auction and I ended up with scratched sensor lol.

Remember a100 d80 and Pentax all are using the same sensor as for Canon they did not use it in the D30 and many were disappointed But canon new the limitations of APS sensors and didn’t up the 8mps.

But rebel is aimed at newbie that only count pixels and compare cameras in that way.

Regards
--
You're welcome to visit my favorite Gallery
http://www.pbase.com/aarif/favorites