Sign in with
Sign up | Sign in
Your question

raw versus jpeg losses quantified

Last response: in Digital Camera
Share
Anonymous
January 31, 2005 2:39:43 AM

Archived from groups: rec.photo.digital (More info?)

Hi all,

In another thread, raw versus jpeg losses have been
discussed. I've quantified jpeg and raw conversion
losses and present my results at:

http://clarkvision.com/imagedetail/raw.versus.jpeg1

Here I show that jpeg losses are a function of intensity
within a pixel, ISO, and camera (related to sensor pixel size).

Roger Clark
Photography at: http://clarkvision.com
Anonymous
January 31, 2005 1:48:23 PM

Archived from groups: rec.photo.digital (More info?)

Roger N. Clark (change username to rnclark) wrote:
> Hi all,
>
> In another thread, raw versus jpeg losses have been
> discussed. I've quantified jpeg and raw conversion
> losses and present my results at:
>
> http://clarkvision.com/imagedetail/raw.versus.jpeg1
>
> Here I show that jpeg losses are a function of intensity
> within a pixel, ISO, and camera (related to sensor pixel size).

Roger,

Thanks again for sharing some most interesting results with us. Whilst
the results are broadly along the lines I would expect, two things
surprise me:

- the JPEG clipping in the Canon 10D
- the lack of benefit of RAW in the S60

I would expect that the 10D issue is just one of the firmware - they could
put in a softer clipping limit for the highlights but chose not to.

Why is the S60 so bad? I thought that it might be an old cheap-and-nasty
camera, but it's actually less that a year old. However, it does have a
5MP 1/1.8" sensor. Perhaps you are seeing the best that the little chip
can really do? You have to ask, then, if these chips can't benefit from
RAW what's the point in adding it to the camera? Just a marketing
gimmick? Perhaps you could also test a 3MP camera with the larger 8.8 x
6.6mm sensor?

Cheers,
David
Anonymous
January 31, 2005 1:48:24 PM

Archived from groups: rec.photo.digital (More info?)

David J Taylor wrote:

> Roger N. Clark (change username to rnclark) wrote:
>
>>Hi all,
>>
>>In another thread, raw versus jpeg losses have been
>>discussed. I've quantified jpeg and raw conversion
>>losses and present my results at:
>>
>>http://clarkvision.com/imagedetail/raw.versus.jpeg1
>>
>>Here I show that jpeg losses are a function of intensity
>>within a pixel, ISO, and camera (related to sensor pixel size).
>
>
> Roger,
>
> Thanks again for sharing some most interesting results with us. Whilst
> the results are broadly along the lines I would expect, two things
> surprise me:
>
> - the JPEG clipping in the Canon 10D
> - the lack of benefit of RAW in the S60
>
> I would expect that the 10D issue is just one of the firmware - they could
> put in a softer clipping limit for the highlights but chose not to.
>
> Why is the S60 so bad? I thought that it might be an old cheap-and-nasty
> camera, but it's actually less that a year old. However, it does have a
> 5MP 1/1.8" sensor. Perhaps you are seeing the best that the little chip
> can really do? You have to ask, then, if these chips can't benefit from
> RAW what's the point in adding it to the camera? Just a marketing
> gimmick? Perhaps you could also test a 3MP camera with the larger 8.8 x
> 6.6mm sensor?
>
> Cheers,
> David
>
>

On the Canon 10D, the clipping must have been a design decision.
It still gets 10 or so stops of dynamic range.

On the S60, it is simply a matter of full well capacity and
read noise.

The S60 has a full well capacity of 11,000 electrons compared to
the 10D of over 44,000. And the S60 needs ISO 50 to reach
that 11,000. At ISO 100 it only gets 5,500 photons.
This will be true of all similar size sensors. Then the S60
has slightly higher equivalent read noise. At ISO 100 the gain
on the 10D is 11.4 electrons/DN but the S60 is 1.5 electrons
per DN. Thus the read noise of 13.6 electrons translates
to a noise DN of 13.6/1.5 = 9, whereas the 10D has noise
DN at ISO 100 = 15.9/11.4 1.4 DN. So the smaller sensor loses
in 2 ways: full well capacity, and noise through lower gain.

(I think of gain as the inverse of the electronics industry:
DN/electron: 10D = 0.088 at ISO 100 compared to S60 = 0.67 DN/electron.)

Roger
Related resources
Can't find your answer ? Ask !
Anonymous
January 31, 2005 7:42:53 PM

Archived from groups: rec.photo.digital (More info?)

Roger N. Clark (change username to rnclark) wrote:
[]

>>> http://clarkvision.com/imagedetail/raw.versus.jpeg1
[]
> The S60 has a full well capacity of 11,000 electrons compared to
> the 10D of over 44,000. And the S60 needs ISO 50 to reach
> that 11,000. At ISO 100 it only gets 5,500 photons.
> This will be true of all similar size sensors. Then the S60
> has slightly higher equivalent read noise. At ISO 100 the gain
> on the 10D is 11.4 electrons/DN but the S60 is 1.5 electrons
> per DN. Thus the read noise of 13.6 electrons translates
> to a noise DN of 13.6/1.5 = 9, whereas the 10D has noise
> DN at ISO 100 = 15.9/11.4 1.4 DN. So the smaller sensor loses
> in 2 ways: full well capacity, and noise through lower gain.
>
> (I think of gain as the inverse of the electronics industry:
> DN/electron: 10D = 0.088 at ISO 100 compared to S60 = 0.67
> DN/electron.)
> Roger

Ah, so if the S60 has an ISO 50 mode, any chance of adding that data to
the figures? It seems a little unfair not to show the camera at its best.

Cheers,
David
Anonymous
January 31, 2005 8:11:24 PM

Archived from groups: rec.photo.digital (More info?)

Roger N. Clark (change username to rnclark) wrote:
> Hi all,
>
> In another thread, raw versus jpeg losses have been
> discussed. I've quantified jpeg and raw conversion
> losses and present my results at:
>
> http://clarkvision.com/imagedetail/raw.versus.jpeg1
>
> Here I show that jpeg losses are a function of intensity
> within a pixel, ISO, and camera (related to sensor pixel size).
>
> Roger Clark
> Photography at: http://clarkvision.com


As usual Roger, you have given us a wealth of solid information
that goes far beyond what we can get from other sources. Thank
you very much for contributing this information and for raising
our level of understanding of the technology.

If you have the time, would you perhaps explain some of the
technical parts to us laymen with insufficient scientific
background to immediately understand them?

Some questions I have are:

1. What is the meaning of the numbers in the scale for Scene
Intensity (Photographic Stops)?

2. When you speak of "full-well capacity", is that the number
of electrons that can accumulate at a single photosite/pixel
on the sensor?

3. What is the "noise due to photon counting statistics", and
why is it the square root of the full-well capacity?

I know these questions indicate my ignorance of some fundamental
issues and I fully understand if it would be too difficult to
explain them. However perhaps you can point us to a website that
explains these and/or similar concepts.

Thanks.

Alan
Anonymous
January 31, 2005 11:13:10 PM

Archived from groups: rec.photo.digital (More info?)

"David J Taylor" <david-taylor@invalid.com> wrote:
>
> Why is the S60 so bad? I thought that it might be an old cheap-and-nasty
> camera, but it's actually less that a year old. However, it does have a
> 5MP 1/1.8" sensor. Perhaps you are seeing the best that the little chip
> can really do?

As I've been ranting for a while now: digital cameras are getting _worse_,
not better. More pixels in less space is simply not a good idea.

> You have to ask, then, if these chips can't benefit from
> RAW what's the point in adding it to the camera? Just a marketing
> gimmick? Perhaps you could also test a 3MP camera with the larger 8.8 x
> 6.6mm sensor?

Was there ever a 3MP 8.8 x 6.6 mm sensor camera? That's the 2/3" sensor, and
I first became aware of that at the 5MP level. Maybe the Sony F505??? Nope,
that was 2MP in a 1/2" sensor.

David J. Littleboy
Tokyo, Japan
Anonymous
January 31, 2005 11:13:11 PM

Archived from groups: rec.photo.digital (More info?)

David J. Littleboy wrote:
> "David J Taylor" <david-taylor@invalid.com> wrote:
>>
>> Why is the S60 so bad? I thought that it might be an old
>> cheap-and-nasty camera, but it's actually less that a year old.
>> However, it does have a 5MP 1/1.8" sensor. Perhaps you are seeing
>> the best that the little chip can really do?
>
> As I've been ranting for a while now: digital cameras are getting
> _worse_, not better. More pixels in less space is simply not a good
> idea.

Well, they offer a range of size and resolution combinations - for me the
5MP level is adequate and I would prefer to see a somewhat larger sensor
size to lower the noise a little.

>> You have to ask, then, if these chips can't benefit from
>> RAW what's the point in adding it to the camera? Just a marketing
>> gimmick? Perhaps you could also test a 3MP camera with the larger
>> 8.8 x
>> 6.6mm sensor?
>
> Was there ever a 3MP 8.8 x 6.6 mm sensor camera? That's the 2/3"
> sensor, and I first became aware of that at the 5MP level. Maybe the
> Sony F505??? Nope, that was 2MP in a 1/2" sensor.

OK, compare whatever is available - like a 2/3" 5MP sensor versus the
1/1.8" sensor in the S60 Roger had to hand. I'm trying to get at whether
any of the smaller than DLSR sensors can benefit from RAW. It seems to be
somewhat against "accepted wisdom".

Cheers,
David
Anonymous
January 31, 2005 11:13:11 PM

Archived from groups: rec.photo.digital (More info?)

On Mon, 31 Jan 2005 20:13:10 +0900, "David J. Littleboy"
<davidjl@gol.com> wrote:

>
>"David J Taylor" <david-taylor@invalid.com> wrote:
>>
>> Why is the S60 so bad? I thought that it might be an old cheap-and-nasty
>> camera, but it's actually less that a year old. However, it does have a
>> 5MP 1/1.8" sensor. Perhaps you are seeing the best that the little chip
>> can really do?
>
>As I've been ranting for a while now: digital cameras are getting _worse_,
>not better. More pixels in less space is simply not a good idea.
>
>> You have to ask, then, if these chips can't benefit from
>> RAW what's the point in adding it to the camera? Just a marketing
>> gimmick? Perhaps you could also test a 3MP camera with the larger 8.8 x
>> 6.6mm sensor?
>
>Was there ever a 3MP 8.8 x 6.6 mm sensor camera? That's the 2/3" sensor, and
>I first became aware of that at the 5MP level. Maybe the Sony F505??? Nope,
>that was 2MP in a 1/2" sensor.
>
>David J. Littleboy
>Tokyo, Japan
>
>
David J. Littleboy,
I don't know if anybody ever made a 3MP digital
camera with a 2/3" CCD but HP's C912 claims 2.24MP with a 2/3" CCD.
Yes I own 1 of them & it's an interesting camera in many ways but it's
general appeal was relatively limited. This camera does exhibit
considerable dynamic range even in jpg format.

As an armature photographer with a modest budget I use some
very simple tools available on the Internet to compare basic noise &
dynamic range of different cameras.

Here is a link to Daves-digicams review of the HP C912:

http://www.steves-digicams.com/c912.html

The picture I like to most in the 1 of the white Marina Cafe':

http://www.steves-digicams.com/c912/samples/IM000186.JP...

The above picture was taken @ ISO 25, yes 25. There is also a
TIFF taken of the same building @ ISO 25.

My overly simplistic use of these pictures is simply to
visually compare the amount of noise in the shaded main entrance area
with the overall exposure of the picture. In general if all things if
all things are equal, the picture with the least noise in the shadow
areas seems the better choice.

Yes I consider far more than this when selecting a digital
camera but this remains an important consideration. Here is the same
picture taken with a Canon A95 @ ISO 50 in jpg:

http://www.steves-digicams.com/2004_reviews/a95/samples...

Over 4 years of advancements separate these 2 cameras & I like
them both but the HP C912 still gets uses & even does some things that
the A95 can't & the reverse is also true.

Hope this helps add some additional information those like
you, with far more knowledge & photo editing tools can use to evaluate
this humble & unusual camera. It, just like my A40 & A60, clearly
illustrate that more MP is no assurance of increased picture quality.
I'd rather a lower resolution but clean 2MP picture over a 5MP or
7.1MP noisier picture in most every case.

Respectfully, DHB


"To announce that there must be no criticism of the President,
or that we are to stand by the President, right or wrong,
is not only unpatriotic and servile, but is morally treasonable
to the American public."--Theodore Roosevelt, May 7, 1918
Anonymous
January 31, 2005 11:13:11 PM

Archived from groups: rec.photo.digital (More info?)

In article <ctl434$sti$1@nnrp.gol.com>, David J. Littleboy
<davidjl@gol.com> wrote:

> Was there ever a 3MP 8.8 x 6.6 mm sensor camera? That's the 2/3" sensor, and
> I first became aware of that at the 5MP level. Maybe the Sony F505??? Nope,
> that was 2MP in a 1/2" sensor.

olympus 2500, 2.5mp
http://www.dpreview.com/reviews/specs/Olympus/oly_c2500...

olympus e10, 4mp
http://www.dpreview.com/reviews/specs/Olympus/oly_e10.a...

olympus e20, 5mp
http://www.dpreview.com/reviews/specs/Olympus/oly_e20.a...
Anonymous
January 31, 2005 11:13:12 PM

Archived from groups: rec.photo.digital (More info?)

"David J Taylor" <david-taylor@invalid.com> wrote in message
news:366jrhF4t904bU1@individual.net...
> David J. Littleboy wrote:
> > "David J Taylor" <david-taylor@invalid.com> wrote:
> >>
> >> Why is the S60 so bad? I thought that it might be an old
> >> cheap-and-nasty camera, but it's actually less that a year old.
> >> However, it does have a 5MP 1/1.8" sensor. Perhaps you are
seeing
> >> the best that the little chip can really do?
> >
> > As I've been ranting for a while now: digital cameras are getting
> > _worse_, not better. More pixels in less space is simply not a
good
> > idea.
>
> Well, they offer a range of size and resolution combinations - for
me the
> 5MP level is adequate and I would prefer to see a somewhat larger
sensor
> size to lower the noise a little.
>
> >> You have to ask, then, if these chips can't benefit from
> >> RAW what's the point in adding it to the camera? Just a
marketing
> >> gimmick? Perhaps you could also test a 3MP camera with the
larger
> >> 8.8 x
> >> 6.6mm sensor?
> >
> > Was there ever a 3MP 8.8 x 6.6 mm sensor camera? That's the 2/3"
> > sensor, and I first became aware of that at the 5MP level. Maybe
the
> > Sony F505??? Nope, that was 2MP in a 1/2" sensor.
>
> OK, compare whatever is available - like a 2/3" 5MP sensor versus
the
> 1/1.8" sensor in the S60 Roger had to hand. I'm trying to get at
whether
> any of the smaller than DLSR sensors can benefit from RAW. It seems
to be
> somewhat against "accepted wisdom".
>
> Cheers,
> David
>
With my previous camera, a Canon S30, pictures produced using RAW were
significantly better than the highest quality jpegs, although after
dabbling with RAW I used jpeg for normal shooting due to time
constraints.

Having now 'moved up' to an S60, I am very dissapointed with the
increased noise from the sensor, although the 5MP and 28mm equivalent
lens are nice to have. Although I haven't tried RAW yet on the S60, I
wouldn't be surprised if there was no particular advantage over jpeg
because of the much noisier pictures.

AR
Anonymous
January 31, 2005 11:13:12 PM

Archived from groups: rec.photo.digital (More info?)

David J Taylor wrote:

> David J. Littleboy wrote:
>
>>"David J Taylor" <david-taylor@invalid.com> wrote:
>>
>>>Why is the S60 so bad? I thought that it might be an old
>>>cheap-and-nasty camera, but it's actually less that a year old.
>>>However, it does have a 5MP 1/1.8" sensor. Perhaps you are seeing
>>>the best that the little chip can really do?
>>
>>As I've been ranting for a while now: digital cameras are getting
>>_worse_, not better. More pixels in less space is simply not a good
>>idea.
>
>
> Well, they offer a range of size and resolution combinations - for me the
> 5MP level is adequate and I would prefer to see a somewhat larger sensor
> size to lower the noise a little.
>
>
>>> You have to ask, then, if these chips can't benefit from
>>>RAW what's the point in adding it to the camera? Just a marketing
>>>gimmick? Perhaps you could also test a 3MP camera with the larger
>>>8.8 x
>>>6.6mm sensor?
>>
>>Was there ever a 3MP 8.8 x 6.6 mm sensor camera? That's the 2/3"
>>sensor, and I first became aware of that at the 5MP level. Maybe the
>>Sony F505??? Nope, that was 2MP in a 1/2" sensor.
>
>
> OK, compare whatever is available - like a 2/3" 5MP sensor versus the
> 1/1.8" sensor in the S60 Roger had to hand. I'm trying to get at whether
> any of the smaller than DLSR sensors can benefit from RAW. It seems to be
> somewhat against "accepted wisdom".

In every case at low ISO, raw is better than jpeg. This is mainly
in the highlights on the small cameras. While there is
variable benefit, there is always benefit. One needs to
then decide for yourself if the added memory space and processing
is worth it to use raw for a particular image.

Roger
Anonymous
January 31, 2005 11:13:13 PM

Archived from groups: rec.photo.digital (More info?)

Roger N. Clark (change username to rnclark) wrote:
[]
> In every case at low ISO, raw is better than jpeg. This is mainly
> in the highlights on the small cameras. While there is
> variable benefit, there is always benefit. One needs to
> then decide for yourself if the added memory space and processing
> is worth it to use raw for a particular image.
>
> Roger

I must confess that noise in the highlights concerns me far less than
noise in the shadows. It's the shadows that you may want to boost to
bring out detail.

I've never used raw or seen the need for it (but I use small sensor
cameras). Your excellent note doesn't yet cause me to change my view. I
would be interested in seeing a result from a 4/3 camera, though.

Cheers,
David
Anonymous
February 1, 2005 1:31:02 AM

Archived from groups: rec.photo.digital (More info?)

David J Taylor wrote:
> Roger N. Clark (change username to rnclark) wrote:
> []
>
>
>>>>http://clarkvision.com/imagedetail/raw.versus.jpeg1
>
> []
>
>>The S60 has a full well capacity of 11,000 electrons compared to
>>the 10D of over 44,000. And the S60 needs ISO 50 to reach
>>that 11,000. At ISO 100 it only gets 5,500 photons.
>>This will be true of all similar size sensors. Then the S60
>>has slightly higher equivalent read noise. At ISO 100 the gain
>>on the 10D is 11.4 electrons/DN but the S60 is 1.5 electrons
>>per DN. Thus the read noise of 13.6 electrons translates
>>to a noise DN of 13.6/1.5 = 9, whereas the 10D has noise
>>DN at ISO 100 = 15.9/11.4 1.4 DN. So the smaller sensor loses
>>in 2 ways: full well capacity, and noise through lower gain.
>>
>>(I think of gain as the inverse of the electronics industry:
>>DN/electron: 10D = 0.088 at ISO 100 compared to S60 = 0.67
>>DN/electron.)
>>Roger
>
>
> Ah, so if the S60 has an ISO 50 mode, any chance of adding that data to
> the figures? It seems a little unfair not to show the camera at its best.
>
> Cheers,
> David
>
>
David,
Yes, I'll add that. Give me a day or so.
Roger
Anonymous
February 1, 2005 1:51:16 AM

Archived from groups: rec.photo.digital (More info?)

Alan Meyer wrote:

> Roger N. Clark (change username to rnclark) wrote:
>
>>Hi all,
>>
>>In another thread, raw versus jpeg losses have been
>>discussed. I've quantified jpeg and raw conversion
>>losses and present my results at:
>>
>>http://clarkvision.com/imagedetail/raw.versus.jpeg1
>>
>>Here I show that jpeg losses are a function of intensity
>>within a pixel, ISO, and camera (related to sensor pixel size).
>>
>>Roger Clark
>>Photography at: http://clarkvision.com
>
>
>
> As usual Roger, you have given us a wealth of solid information
> that goes far beyond what we can get from other sources. Thank
> you very much for contributing this information and for raising
> our level of understanding of the technology.

Thanks.
>
> If you have the time, would you perhaps explain some of the
> technical parts to us laymen with insufficient scientific
> background to immediately understand them?
>
> Some questions I have are:
>
> 1. What is the meaning of the numbers in the scale for Scene
> Intensity (Photographic Stops)?

That is the linear output from the sensor,
expressed in stops.
>
> 2. When you speak of "full-well capacity", is that the number
> of electrons that can accumulate at a single photosite/pixel
> on the sensor?

Yes.
>
> 3. What is the "noise due to photon counting statistics", and
> why is it the square root of the full-well capacity?

The fundamental error in measuring a photon signal is the square
root of the number of photons counted, Poisson Statistics.
The maximum number of photons one can count with a sensor
is the max number of electrons in that can be held in the well.
There is one electron per photon. If one fills the pixel
well with 40,000 electrons, then the noise in the signal is
square root 40,000. So whatever the signal is, the error
(noise) is square root of the number of electrons (photons).
The more photons counted, the higher the signal-to-noise.
The signal-to-noise = # photons/square root(# photons)
= square root(# photons)
In the shadows in an image, one may get only a few hundred
photons, so the noise is square root of those few hundred.
Photons Noise signal-to-noise
9 3 3
100 10 10
900 30 30
10000 100 100
40000 200 200

> I know these questions indicate my ignorance of some fundamental
> issues and I fully understand if it would be too difficult to
> explain them. However perhaps you can point us to a website that
> explains these and/or similar concepts.

Sometimes it is me not being clear in my writing, so I welcome
questions when I need to clear things up.
I'll see if I can find some references (web pages) and
add them to the page.

Roger
Anonymous
February 1, 2005 11:00:55 AM

Archived from groups: rec.photo.digital (More info?)

Roger N. Clark (change username to rnclark) wrote:
> ...
> Sometimes it is me not being clear in my writing, so I welcome
> questions when I need to clear things up.
> I'll see if I can find some references (web pages) and
> add them to the page.
> ...

Thanks very much for the explanations and for any references you
can provide.

One of the problems in reading about digital photography
technology is that it's easy to find glossy material that
introduces the complete beginner to high level concepts, and it's
possible to find some materials written by experts for experts,
but there's too little material that bridges the gap, bringing
the reader up from the beginner level to where he can at least
begin to understand the underlying science.

If I were the Secretary of Education in the U.S., I would fund a
major effort to put scientific and technical education on the
web, accessible to all. Right now, we are all dependent on a few
unpaid volunteers with special interests such as yourself for
information.

Thanks again.

Alan
Anonymous
February 1, 2005 11:07:22 AM

Archived from groups: rec.photo.digital (More info?)

Roger N. Clark (change username to rnclark) wrote:
> ...
> Sometimes it is me not being clear in my writing, so I welcome
> questions when I need to clear things up.
> I'll see if I can find some references (web pages) and
> add them to the page.
> ...

Thanks very much for the explanations and for any references you
can provide.

One of the problems in reading about digital photography
technology is that it's easy to find glossy material that
introduces the complete beginner to high level concepts, and it's
possible to find some materials written by experts for experts,
but there's too little material that bridges the gap, bringing
the reader up from the beginner level to where he can at least
begin to understand the underlying science.

If I were the Secretary of Education in the U.S., I would fund a
major effort to put scientific and technical education on the
web, accessible to all. Right now, we are all dependent on a few
unpaid volunteers with special interests such as yourself for
information.

Thanks again.

Alan
Anonymous
February 1, 2005 12:39:10 PM

Archived from groups: rec.photo.digital (More info?)

Clarks site was incredible!
One is always looking for ways to un-obtrusivly sharpen digitial photos.
Clark and other sharpening sites talk about the "Richardson-Lucy Algorithm".
Is this a standlone WindowsXP program, a plugin for PhotoShop or is it a
command line program? In a Google search, I found that astronomers mainly
use it to sharpen their shots done with a telescope, and I think Clark had
some examples of wildlife pictures he sharpened with the RLA. Also he
discussed multiple unsharp masks at different settings that came close to
the RLA.
However in searching I could not find THE actual program. Does anyone know
where this can be found/purchased?
Yours,
Tom
"Alan Meyer" <ameyer2@yahoo.com> wrote in message
news:1107220284.898221.114510@z14g2000cwz.googlegroups.com...
> Roger N. Clark (change username to rnclark) wrote:
> > Hi all,
> >
> > In another thread, raw versus jpeg losses have been
> > discussed. I've quantified jpeg and raw conversion
> > losses and present my results at:
> >
> > http://clarkvision.com/imagedetail/raw.versus.jpeg1
> >
> > Here I show that jpeg losses are a function of intensity
> > within a pixel, ISO, and camera (related to sensor pixel size).
> >
> > Roger Clark
> > Photography at: http://clarkvision.com
>
>
> As usual Roger, you have given us a wealth of solid information
> that goes far beyond what we can get from other sources. Thank
> you very much for contributing this information and for raising
> our level of understanding of the technology.
>
> If you have the time, would you perhaps explain some of the
> technical parts to us laymen with insufficient scientific
> background to immediately understand them?
>
> Some questions I have are:
>
> 1. What is the meaning of the numbers in the scale for Scene
> Intensity (Photographic Stops)?
>
> 2. When you speak of "full-well capacity", is that the number
> of electrons that can accumulate at a single photosite/pixel
> on the sensor?
>
> 3. What is the "noise due to photon counting statistics", and
> why is it the square root of the full-well capacity?
>
> I know these questions indicate my ignorance of some fundamental
> issues and I fully understand if it would be too difficult to
> explain them. However perhaps you can point us to a website that
> explains these and/or similar concepts.
>
> Thanks.
>
> Alan
>
Anonymous
February 1, 2005 2:05:04 PM

Archived from groups: rec.photo.digital (More info?)

Tom Ellliott wrote:

> However in searching I could not find THE actual program. Does anyone
know
> where this can be found/purchased?

??? Clark offers links to the software he uses.

http://clarkvision.com/imagedetail/image-restoration1; look at the
bottom of the page for his references.
Anonymous
February 1, 2005 2:09:21 PM

Archived from groups: rec.photo.digital (More info?)

Roger N. Clark (change username to rnclark) wrote:
[]
>> Ah, so if the S60 has an ISO 50 mode, any chance of adding that data
>> to the figures? It seems a little unfair not to show the camera at
>> its best. Cheers,
>> David
>>
>>
> David,
> Yes, I'll add that. Give me a day or so.
> Roger

Thanks, Roger.
David
Anonymous
February 1, 2005 4:47:48 PM

Archived from groups: rec.photo.digital (More info?)

"Roger N. Clark (change username to rnclark)" <username@qwest.net> wrote:
> Hi all,

> In another thread, raw versus jpeg losses have been
> discussed. I've quantified jpeg and raw conversion
> losses and present my results at:

> http://clarkvision.com/imagedetail/raw.versus.jpeg1

> Here I show that jpeg losses are a function of intensity
> within a pixel, ISO, and camera (related to sensor pixel size).

There's another angle to consider: the losses when going from the raw
colour space to the working space. A little while ago, one poster to
this group complained about a atill life he had photographed with a
digital camera. The oranges were the same colour as the tomatoes! It
looks like some cameras use fairly crude colorimetic rendering when
going from raw RGB->working RGB. Of course, this is a problem for
TIFF too, not just JPEG. But the conversion from raw->TIFF cannot be
said truly to be lossless, unless you make the transformation into
wide gamut RGB.

Andrew.
Anonymous
February 1, 2005 5:07:16 PM

Archived from groups: rec.photo.digital (More info?)

andrew29@littlepinkcloud.invalid wrote:
[]
> There's another angle to consider: the losses when going from the raw
> colour space to the working space. A little while ago, one poster to
> this group complained about a atill life he had photographed with a
> digital camera. The oranges were the same colour as the tomatoes! It
> looks like some cameras use fairly crude colorimetic rendering when
> going from raw RGB->working RGB. Of course, this is a problem for
> TIFF too, not just JPEG. But the conversion from raw->TIFF cannot be
> said truly to be lossless, unless you make the transformation into
> wide gamut RGB.
>
> Andrew.

That is an excellent point, Andrew. I recall that in tests of "how many
bits do you need for digital video" whilst 6 bits were almost enough for
monochrome, the colour sensitivity of the eye meant that 8 bits of RGB
were only just enough for colour. Again, I would expect this to be more
important at the darker end of the JPEG range than when photographing
(presumably) fairly bright tomatoes or oranges.

Cheers,
David
Anonymous
February 1, 2005 6:11:53 PM

Archived from groups: rec.photo.digital (More info?)

Sorry about the double post.

Google reported that my posting was lost in a server
error and told me to try again, so I did. But it
appears that both postings went through.

Alan
Anonymous
February 2, 2005 12:05:04 AM

Archived from groups: rec.photo.digital (More info?)

David J Taylor wrote:

> Roger N. Clark (change username to rnclark) wrote:
> []
>
>>>Ah, so if the S60 has an ISO 50 mode, any chance of adding that data
>>>to the figures? It seems a little unfair not to show the camera at
>>>its best. Cheers,
>>>David
>>>
>>>
>>
>>David,
>>Yes, I'll add that. Give me a day or so.
>>Roger
>
>
> Thanks, Roger.
> David
>
>
Done!
Anonymous
February 2, 2005 5:27:40 AM

Archived from groups: rec.photo.digital (More info?)

"Roger N. Clark (change username to rnclark)" <username@qwest.net>
wrote in message news:41FF18D4.6010204@qwest.net...
SNIP
> The fundamental error in measuring a photon signal is the square
> root of the number of photons counted, Poisson Statistics.

And to prevent trollish responses from certain (John Pike like) types,
Roger is deliberately simplifying here to clarify a (by using a
simplified (large quantity)) characteristic of the Poisson
distribution. At large quantities, the (asymmetric) Poisson
distribution approaches a Gaussian distribution.

> The maximum number of photons one can count with a sensor is the max
> number of electrons in that can be held in the well. There is one
> electron per photon.

Which again is a slight simplification for clarity, if the sensor's
wavelength dependent Quantum Detection Efficiency at capture is
disregarded.

> If one fills the pixel
> well with 40,000 electrons, then the noise in the signal is
> square root 40,000. So whatever the signal is, the error
> (noise) is square root of the number of electrons (photons).
> The more photons counted, the higher the signal-to-noise.
> The signal-to-noise = # photons/square root(# photons)
> = square root(# photons)
> In the shadows in an image, one may get only a few hundred
> photons, so the noise is square root of those few hundred.
> Photons Noise signal-to-noise
> 9 3 3
> 100 10 10
> 900 30 30
> 10000 100 100
> 40000 200 200
>
>> I know these questions indicate my ignorance of some fundamental
>> issues and I fully understand if it would be too difficult to
>> explain them. However perhaps you can point us to a website that
>> explains these and/or similar concepts.
>
> Sometimes it is me not being clear in my writing, so I welcome
> questions when I need to clear things up.
> I'll see if I can find some references (web pages) and
> add them to the page.

These may be useful:
http://www.roperscientific.com/pdfs/technotes/snr.pdf
http://www.ccd.com/ccd101.html , http://www.ccd.com/ccd111.html or
http://www.ccd.com/ccdu.html in general.

Bart
Anonymous
February 2, 2005 5:27:41 AM

Archived from groups: rec.photo.digital (More info?)

Bart van der Wolf wrote:

>
> "Roger N. Clark (change username to rnclark)" <username@qwest.net> wrote
> in message news:41FF18D4.6010204@qwest.net...
> SNIP
>
>> The fundamental error in measuring a photon signal is the square root
>> of the number of photons counted, Poisson Statistics.
>
>
> And to prevent trollish responses from certain (John Pike like) types,
> Roger is deliberately simplifying here to clarify a (by using a
> simplified (large quantity)) characteristic of the Poisson distribution.
> At large quantities, the (asymmetric) Poisson distribution approaches a
> Gaussian distribution.

And at low numbers of photons, the noise is dominated by read noise,
so this is a good approximation that seems to be in pretty much
universal use, working well over the entire range of sensor data.

>> The maximum number of photons one can count with a sensor is the max
>> number of electrons in that can be held in the well. There is one
>> electron per photon.
>
>
> Which again is a slight simplification for clarity, if the sensor's
> wavelength dependent Quantum Detection Efficiency at capture is
> disregarded.

Actually, it is exact. I meant one electron per photon that gets
converted to an electron and stored in the well. That is what
we see out of the chip, so it doesn't matter what the quantum
efficiency is in this regard.

I've added the ISO 50 curve to the S60 plot, added a section on
equations used, and references. The ISO 50 on the S60 camera
shows a greater jpeg loss, making raw images a greater difference,
as expected.

The next question is inter comparison of the three cameras:
that will show the differences between pro DSLR, prosumer DSLR,
and consumer P&S cameras.

Roger
Anonymous
February 2, 2005 5:57:54 AM

Archived from groups: rec.photo.digital (More info?)

"Tom Ellliott" <1stroke@bellsouth.net> wrote in message
news:avMLd.4814$3W3.4399@bignews4.bellsouth.net...
> Clarks site was incredible!
> One is always looking for ways to un-obtrusivly sharpen digitial
> photos. Clark and other sharpening sites talk about the "Richardson-
> Lucy Algorithm".
> Is this a standlone WindowsXP program, a plugin for PhotoShop or is
> it a
> command line program?

It's an image processing technique/algorithm that generally restores
image resolution (assuming a Poisson distibution of luminance) for
high S/N and (if applied adaptively) also for low S/N signals. In fact
it strikes an an average usability between High and Low signal to
noise images. It is based on statistical properties of (image) data.

> In a Google search, I found that astronomers mainly use it
> to sharpen their shots done with a telescope, and I think
> Clark had some examples of wildlife pictures he sharpened
> with the RLA. Also he discussed multiple unsharp masks
> at different settings that came close to the RLA.
>
> However in searching I could not find THE actual program.
>Does anyone know where this can be found/purchased?

It is usually a part of (mainly) astronomical image processing
software, some of which can be daunting to use (or in cost, if only
for this function) without sufficient background knowledge or work
volume.
It is a processing expensive function, so less useful for the
impatient or processor speed and computer memory challenged ...

Bart
Anonymous
February 2, 2005 6:11:38 AM

Archived from groups: rec.photo.digital (More info?)

<andrew29@littlepinkcloud.invalid> wrote in message
news:10vv24488tqmr21@news.supernews.com...
SNIP
> But the conversion from raw->TIFF cannot be said truly to be
> lossless, unless you make the transformation into wide
> gamut RGB.

True, Capture device space will/should often be of wider gamut than
the output space, which is why a profiled and a colormanaged workflow
is beneficial for optimal color reproduction. Rendering intent will
also be important if an appropriately wide capture gamut color space
is used.

Bart
Anonymous
February 2, 2005 4:43:32 PM

Archived from groups: rec.photo.digital (More info?)

Bart van der Wolf <bvdwolf@no.spam> wrote:

> <andrew29@littlepinkcloud.invalid> wrote in message
> news:10vv24488tqmr21@news.supernews.com...
> SNIP
>> But the conversion from raw->TIFF cannot be said truly to be
>> lossless, unless you make the transformation into wide
>> gamut RGB.

> True, Capture device space will/should often be of wider gamut than
> the output space, which is why a profiled and a colormanaged workflow
> is beneficial for optimal color reproduction. Rendering intent will
> also be important if an appropriately wide capture gamut color space
> is used.

The trouble is, only the manufacturers know what their cameras are
doing when creating in-camera JPEGs. From what I've seen, I'm
guessing that it's something like relative colorimetric rendering,
where only the white point is moved and out of gamut colours are
clipped.

Andrew.
Anonymous
February 2, 2005 5:15:18 PM

Archived from groups: rec.photo.digital (More info?)

This is a great discussion, and many thanks to Roger, especially as I
have S60!

My feeling from much use of the camera is that at ISO50 I get very good
results, as long as my subject matter does not have too great a dynamic
range. ISO 200 is OK if you need it, because ISO400 gives waaayyyy too
much noise!

But, I bought it because among other things it has RAW mode, which
means I can use my intelligence to convert to a useable image.

And that is the point I want to make from this post. RAW allows the
user to control their own settings, not whatever algorithm the camera
designer thought you might need.

I would also be interested to know how the camera chooses the ISO when
in auto mode.

DB
Anonymous
February 2, 2005 7:37:11 PM

Archived from groups: rec.photo.digital (More info?)

On Tue, 01 Feb 2005 21:00:10 -0700, "Roger N. Clark (change username
to rnclark)" <username@qwest.net> wrote:

>Bart van der Wolf wrote:
>
>>
>> "Roger N. Clark (change username to rnclark)" <username@qwest.net> wrote
>> in message news:41FF18D4.6010204@qwest.net...
>> SNIP
>>
>>> The fundamental error in measuring a photon signal is the square root
>>> of the number of photons counted, Poisson Statistics.
>>
>>
>> And to prevent trollish responses from certain (John Pike like) types,
>> Roger is deliberately simplifying here to clarify a (by using a
>> simplified (large quantity)) characteristic of the Poisson distribution.
>> At large quantities, the (asymmetric) Poisson distribution approaches a
>> Gaussian distribution.
>
>And at low numbers of photons, the noise is dominated by read noise,
>so this is a good approximation that seems to be in pretty much
>universal use, working well over the entire range of sensor data.
>
>>> The maximum number of photons one can count with a sensor is the max
>>> number of electrons in that can be held in the well. There is one
>>> electron per photon.
>>
>>
>> Which again is a slight simplification for clarity, if the sensor's
>> wavelength dependent Quantum Detection Efficiency at capture is
>> disregarded.
>
>Actually, it is exact. I meant one electron per photon that gets
>converted to an electron and stored in the well. That is what
>we see out of the chip, so it doesn't matter what the quantum
>efficiency is in this regard.
>
>I've added the ISO 50 curve to the S60 plot, added a section on
>equations used, and references. The ISO 50 on the S60 camera
>shows a greater jpeg loss, making raw images a greater difference,
>as expected.
>
>The next question is inter comparison of the three cameras:
>that will show the differences between pro DSLR, prosumer DSLR,
>and consumer P&S cameras.
>
>Roger

Roger,
your participation in this news group & wealth of information
on your web site is overwhelming to me at times, but I learn what I
can & hopefully improve my relatively modest abilities as an amateur
digital photographer thanks to people like you.

With that said & at the risk of illustrating my ignorance,
might I offer both an observation & a question?

A jpg compression algorithm does not operate on a fixed
compression ratio, it's influenced by picture "complexity" or "detail"
which is why some pictures compress much more than others with the
same setting being used. This is also effected by ISO settings
because the jpg algorithm has increased difficulty separating "digital
noise" from "detail", hence it's ability to compress a noisier, high
ISO image decreases. This can be seen on many digital cameras simply
by looking at the estimated number pictures remaining on a given
memory card drop when the user increases the ISO setting.

Maybe I am looking at all of this too simplistically but,
could it be that the increased level of compression at lower ISO
setting is a major contributor to the reduction in dynamic range at
lower ISO settings? D-SLR's should do much better because their S/N
ration is much higher due to the larger sensors used in them.

OK, now that I have illustrated my ignorance, I will sit back
& hope that those with much more knowledge than I, will politely
inform me as to what I may be missing & where I can go to try to learn
more in this area.

As always, thanks in advance for even reading my post & for
any advice you & or others may have to offer.

Respectfully, DHB

"To announce that there must be no criticism of the President,
or that we are to stand by the President, right or wrong,
is not only unpatriotic and servile, but is morally treasonable
to the American public."--Theodore Roosevelt, May 7, 1918
Anonymous
February 3, 2005 4:38:58 PM

Archived from groups: rec.photo.digital (More info?)

"DHB" <yoda2k@verizon.net> wrote in message
news:k5u101tu1a6upbi9h28s35c5uctnn9q7f9@4ax.com...
SNIP
> A jpg compression algorithm does not operate on a
> fixed compression ratio, it's influenced by picture
> "complexity" or "detail" which is why some pictures
> compress much more than others with the same setting
> being used. This is also effected by ISO settings
> because the jpg algorithm has increased difficulty
> separating "digital noise" from "detail", hence it's ability
> to compress a noisier, high ISO image decreases.
> This can be seen on many digital cameras simply by
> looking at the estimated number pictures remaining on a
> given memory card drop when the user increases the
> ISO setting.

Correct so far. Just a small nuance, the JPEG algorithm doesn't
attempt to separate noise from detail. It is clueless in that regard,
all it sees is blocks of pixels and bit patterns.

> Maybe I am looking at all of this too simplistically but,
> could it be that the increased level of compression at
> lower ISO setting is a major contributor to the reduction
> in dynamic range at lower ISO settings?

There is no "increased level of compression", the image just happens
to compress better with the same compressor settings. Due to the
reduced noise at lower ISOs, even a lossless compression would be able
to produce a smaller file. Besides, in essence the core of the
compression is lossless, however the amount of data to compress is
"lossy" reduced (elimination of highest spatial frequencies, esp.
chroma) before compression. The compression "quality" setting
determines how much is thrown away. In its highest setting, very
little is lost but the file is less suited for further processing
(e.g. sharpening) as it may increase visibility of compression
artifacts.

Bart
Anonymous
February 3, 2005 9:53:56 PM

Archived from groups: rec.photo.digital (More info?)

On Thu, 3 Feb 2005 13:38:58 +0100, "Bart van der Wolf"
<bvdwolf@no.spam> wrote:

>
>"DHB" <yoda2k@verizon.net> wrote in message
>news:k5u101tu1a6upbi9h28s35c5uctnn9q7f9@4ax.com...
>SNIP
>> A jpg compression algorithm does not operate on a
>> fixed compression ratio, it's influenced by picture
>> "complexity" or "detail" which is why some pictures
>> compress much more than others with the same setting
>> being used. This is also effected by ISO settings
>> because the jpg algorithm has increased difficulty
>> separating "digital noise" from "detail", hence it's ability
>> to compress a noisier, high ISO image decreases.
>> This can be seen on many digital cameras simply by
>> looking at the estimated number pictures remaining on a
>> given memory card drop when the user increases the
>> ISO setting.
>
>Correct so far. Just a small nuance, the JPEG algorithm doesn't
>attempt to separate noise from detail. It is clueless in that regard,
>all it sees is blocks of pixels and bit patterns.
>
>> Maybe I am looking at all of this too simplistically but,
>> could it be that the increased level of compression at
>> lower ISO setting is a major contributor to the reduction
>> in dynamic range at lower ISO settings?
>
>There is no "increased level of compression", the image just happens
>to compress better with the same compressor settings. Due to the
>reduced noise at lower ISOs, even a lossless compression would be able
>to produce a smaller file. Besides, in essence the core of the
>compression is lossless, however the amount of data to compress is
>"lossy" reduced (elimination of highest spatial frequencies, esp.
>chroma) before compression. The compression "quality" setting
>determines how much is thrown away. In its highest setting, very
>little is lost but the file is less suited for further processing
>(e.g. sharpening) as it may increase visibility of compression
>artifacts.
>
>Bart

Bart,
thanks for taking the time to explain that in greater detail,
apparently I did oversimplify it but in part because I did not
understand it all that well myself. Now I have a better understanding
& additional interest to learn more about it.

Respectfully, DHB


"To announce that there must be no criticism of the President,
or that we are to stand by the President, right or wrong,
is not only unpatriotic and servile, but is morally treasonable
to the American public."--Theodore Roosevelt, May 7, 1918
!