707 in-camera Sharpening - Low Pass Filters

Simon Glynn · Oct 21, 2002

Hi All,

SUMMARY OF THIS POST ------------------------------------------------------

I discuss in-camera sharpening on the 707 and wonder whether level -2 sharpening is actually the least sharpened image and whether you should move toward performing sharpening in post-processing as a result. I currently run with level 0 sharpening, but am finding certain scenes have significant haloing which I would like to have greater control over.

---------------------------------------------------------------------------------------

The following two articles.

http://www.robgalbraith.com/diginews/2002-10/2002_10_20_eos1ds.html

http://www.microsoft.com/windowsxp/digitalphotography/learnmore/fixit/sharpening.asp

are in realtion to, or have a section relating to in-camera sharpening. They identify that the inclusion of an optical low pass filter has the effect of reducing undesirable digital artifacts in images, but has the disadvantage of reducing the sharpness of the final image. They indicate that an optical low pass filter is likely to be present in most digital cameras so I'll make the assumption that the 707 has this filter, based on its ability to alter image sharpness in-camera.

The primary message in both of the articles, was that in-camera sharpening is either not as powerful or not as flexible as post-processing sharpness, and that leads to my question.

QUESTION: Is there a reason that the 707 would have the ability to reduce sharpness of an image, beyond that which is natively produced at the CCD (ignoring the RAW to TIFF/JPG conversion)? In other words Is there reason to believe that -2 sharpening is a deliberate bluring of the image, or alternatively is it the least sharpened version of the captured image

I understand from previous posts, that there are those who find 0 level sharpening to produce an image that appears to be most representative of the captured scene, however my own experience on high contrast subjects generates photos with significant "halo" sharpening artifacts when using 0 level sharpening.

I will hypothesise the following, and would be keen to hear feedback in relation to the following and my question..........

HYPOTHESIS

If in-camera sharpening operates on the RAW CCD data (i.e. on the 5 million pixels, most of which are Green, some Red and some Blue) then it may not be possible to match in-camera sharpening via post processing (Why?, well we don't have access to the RAW data and hence can't replicate or improve on the in-camera processing prior to the Rpixel Gpixel Bpixel data being converted into an RGB per pixel image).

If in-camera sharpening occurs AFTER the RAW -> combined image conversion stage, then if minimal in-camera sharpening is applied, it should be possible to recreate, or improve on in-camera sharping via post processing (Why? in-camera sharpening has no adjustable parameters, e.g. radius, threshhold, amount, and may be restricted based on camera CPU processing power)

(I'm hoping this is the case) -2 in-camera sharpening is the least sharpened image. It's blurred appearance is due to the low pass filter in the optical path, and it should be possible to achieve or exceed the sharpness achieved using 0 level in-camera sharpening. ..... If I am incorrect and -2 is actually a BLUR filter applied to the image, then it will not be possible to achieve the same sharpness as level 0, as image data will have been lost

Thoughts, and experimental results welcomed. I will try and find appropriate subjects to perform the testing on myself.

Regards
Simon Glynn

Charles C Lloyd · Oct 21, 2002

Simon Glynn said:
(I'm hoping this is the case) -2 in-camera sharpening is the

least sharpened image. It's blurred appearance is due to the low
pass filter in the optical path, and it should be possible to
achieve or exceed the sharpness achieved using 0 level in-camera
sharpening. ..... If I am incorrect and -2 is actually a BLUR
filter applied to the image, then it will not be possible to
achieve the same sharpness as level 0, as image data will have been
lost

I don't have any value to add to this discussion, but am interested in the results. Let me suggest that adding a blur would be an "effect" more appropriate for the P.effect menu option. (Does anyone want that sort of thing done in-camera? Well, anyone besides a store clerk.) One would hope that they didn't leak their P.effects into legitimate camera settings like sharpness.

I hope you're right about the -2 actually being 0 and zero is really 3. In other words, 3 is really the default setting and they just call it 0. Perhaps they consider 0 the level that restores the sharpness introduced by the lowpass optical filter.

Charles.
------
The other day I went to... no wait, that was someone else.
Sony 707 (Since Oct 14, 2002)
Canon S100 (Since Feb 2001)

Simon Glynn · Oct 21, 2002

Charles C Lloyd said:
I hope you're right about the -2 actually being 0 and zero is
really 3. In other words, 3 is really the default setting and they
just call it 0. Perhaps they consider 0 the level that restores
the sharpness introduced by the lowpass optical filter.

Charles, you've summed it up nicely. If results point towards 0 being a restoration of sharpness to pre low-pass filter sharpness, then it should be a piece of cake to improve a -2 sharpness image to exceed the quality of a 0 sharpness image via post processing simply because post processing sharpness can be adjusted to specifically match the characteristics of the captured scene.

If I manage to create some results before another user, I'll try and put them into this thread.

Simon

Hal Fisher · Oct 21, 2002

You might be on to something. If the in-camera sharpness processing was so great we would be using +1 or 2 more then an unsharp filters. I'll have to try a bunch.

Hal

Simon Glynn said:
Charles C Lloyd said:

I hope you're right about the -2 actually being 0 and zero is
really 3. In other words, 3 is really the default setting and they
just call it 0. Perhaps they consider 0 the level that restores
the sharpness introduced by the lowpass optical filter.

Click to expand...

Charles, you've summed it up nicely. If results point towards 0
being a restoration of sharpness to pre low-pass filter sharpness,
then it should be a piece of cake to improve a -2 sharpness image
to exceed the quality of a 0 sharpness image via post processing
simply because post processing sharpness can be adjusted to
specifically match the characteristics of the captured scene.

If I manage to create some results before another user, I'll try
and put them into this thread.

Simon

Todd Moore · Oct 21, 2002

Great post, thanks for the info as you make great sense with this!. I will be playing with this!

Simon Glynn said:
Hi All,

SUMMARY OF THIS POST
------------------------------------------------------
I discuss in-camera sharpening on the 707 and wonder whether level
-2 sharpening is actually the least sharpened image and whether you
should move toward performing sharpening in post-processing as a
result. I currently run with level 0 sharpening, but am finding
certain scenes have significant haloing which I would like to have
greater control over.

---------------------------------------------------------------------------------------

The following two articles.

http://www.robgalbraith.com/diginews/2002-10/2002_10_20_eos1ds.html

http://www.microsoft.com/windowsxp/digitalphotography/learnmore/fixit/sharpening.asp

are in realtion to, or have a section relating to in-camera
sharpening. They identify that the inclusion of an optical low pass
filter has the effect of reducing undesirable digital artifacts in
images, but has the disadvantage of reducing the sharpness of the
final image. They indicate that an optical low pass filter is
likely to be present in most digital cameras so I'll make the
assumption that the 707 has this filter, based on its ability to
alter image sharpness in-camera.

The primary message in both of the articles, was that in-camera
sharpening is either not as powerful or not as flexible as
post-processing sharpness, and that leads to my question.

QUESTION: Is there a reason that the 707 would have the ability to
reduce sharpness of an image, beyond that which is natively
produced at the CCD (ignoring the RAW to TIFF/JPG conversion)? In
other words Is there reason to believe that -2 sharpening is a
deliberate bluring of the image, or alternatively is it the least
sharpened version of the captured image

I understand from previous posts, that there are those who find 0
level sharpening to produce an image that appears to be most
representative of the captured scene, however my own experience on
high contrast subjects generates photos with significant "halo"
sharpening artifacts when using 0 level sharpening.

I will hypothesise the following, and would be keen to hear
feedback in relation to the following and my question..........

HYPOTHESIS

If in-camera sharpening operates on the RAW CCD data (i.e. on the

5 million pixels, most of which are Green, some Red and some Blue)
then it may not be possible to match in-camera sharpening via post
processing (Why?, well we don't have access to the RAW data and
hence can't replicate or improve on the in-camera processing prior
to the Rpixel Gpixel Bpixel data being converted into an RGB per
pixel image).

If in-camera sharpening occurs AFTER the RAW -> combined image

conversion stage, then if minimal in-camera sharpening is applied,
it should be possible to recreate, or improve on in-camera sharping
via post processing (Why? in-camera sharpening has no adjustable
parameters, e.g. radius, threshhold, amount, and may be restricted
based on camera CPU processing power)

(I'm hoping this is the case) -2 in-camera sharpening is the

least sharpened image. It's blurred appearance is due to the low
pass filter in the optical path, and it should be possible to
achieve or exceed the sharpness achieved using 0 level in-camera
sharpening. ..... If I am incorrect and -2 is actually a BLUR
filter applied to the image, then it will not be possible to
achieve the same sharpness as level 0, as image data will have been
lost

Thoughts, and experimental results welcomed. I will try and find
appropriate subjects to perform the testing on myself.

Regards
Simon Glynn

Charles C Lloyd · Oct 22, 2002

Let me be the first to throw out some experimental results from my week-old 707.

First, let me say that, even though I have very limited data to back up my claim, I believe it is best to set the in-camera sharpening to -2. I believe the initial hypothesis is correct that -2 is really 0 sharpening and does not introduce a blur.

I took 5 shots of the same Diet Coke box from about 2 meters with a full 5x zoom. I used sharpening values of:

-2, -1, 0, +1, and +2

I found that -2 had very low noise and was certainly the least sharp of the bunch, while +2 had the most noise and was the sharpest; there were the expected gradients of sharpness and noise in between. All just as anticipated.

So to me the interesting question is can the -2 image be sharpened to at least as good as the +2 image and have the end result contain less noise. I claim I was able to do this and have included links below to back it up.

Caveat: The image I shot was not the sort of thing one would normally shoot, so I can't say what would happen with a real image of a furry animal, for example. I selected the bar code off the box because it was easiest to see the effects of sharpening and had some fairly noisy flat colored region next to it.

The following three images are full crops (no resizing). The second one is the only one I processed in PS Elements. As a beginner PS guy, I think I may have gone overboard a bit on the USM, but these were the recommended starting points from Adobe.

In-camera Sharpening -2 Before PS Element Unsharp Mask:

Same as above, after PS Elements Unsharp Mask (150% 2 pixels/0 threshold). Notice how the noise in the gray area is quite a bit less than the corresponding area of the next shot:

In-camera Sharpening +2:

Feel free to point out the weaknesses/flaws in this test. For my money, I'm going to start shooting at -2 and plan on doing USM (as I always do anyway) and see how it goes from there.

Charles.
------
The other day I went to... no wait, that was someone else.
Sony 707 (Since Oct 14, 2002)
Canon S100 (Since Feb 2001)

Brad Morris · Oct 22, 2002

Charles,

I suggest that you try USM 350-500%, 0.3 radius and between 0-3 threshold depending on how noisy the original image is.

The threshold setting tells PS how contrasy the differences have to be before sharpening is applied. You might find you get better results

It is also My understanding that -2 is close to no sharpening , 0 is bringing the image up to what Sony things it should be without the Low Pass filter

Cheers

Brad

Charles C Lloyd said:
Let me be the first to throw out some experimental results from my
week-old 707.

First, let me say that, even though I have very limited data to
back up my claim, I believe it is best to set the in-camera
sharpening to -2. I believe the initial hypothesis is correct that
-2 is really 0 sharpening and does not introduce a blur.

I took 5 shots of the same Diet Coke box from about 2 meters with a
full 5x zoom. I used sharpening values of:

-2, -1, 0, +1, and +2

I found that -2 had very low noise and was certainly the least
sharp of the bunch, while +2 had the most noise and was the
sharpest; there were the expected gradients of sharpness and noise
in between. All just as anticipated.

So to me the interesting question is can the -2 image be sharpened
to at least as good as the +2 image and have the end result contain
less noise. I claim I was able to do this and have included links
below to back it up.

Caveat: The image I shot was not the sort of thing one would
normally shoot, so I can't say what would happen with a real image
of a furry animal, for example. I selected the bar code off the
box because it was easiest to see the effects of sharpening and had
some fairly noisy flat colored region next to it.

The following three images are full crops (no resizing). The
second one is the only one I processed in PS Elements. As a
beginner PS guy, I think I may have gone overboard a bit on the
USM, but these were the recommended starting points from Adobe.

In-camera Sharpening -2 Before PS Element Unsharp Mask:

Same as above, after PS Elements Unsharp Mask (150% 2 pixels/0
threshold). Notice how the noise in the gray area is quite a bit
less than the corresponding area of the next shot:

In-camera Sharpening +2:

Feel free to point out the weaknesses/flaws in this test. For my
money, I'm going to start shooting at -2 and plan on doing USM (as
I always do anyway) and see how it goes from there.

Charles.
------
The other day I went to... no wait, that was someone else.
Sony 707 (Since Oct 14, 2002)
Canon S100 (Since Feb 2001)

--
DCS-F707, Nikon CP 950

Gordon W · Oct 22, 2002

Simon Glynn said:
Hi All,

SUMMARY OF THIS POST
------------------------------------------------------
I discuss in-camera sharpening on the 707 and wonder whether level
-2 sharpening is actually the least sharpened image and whether you
should move toward performing sharpening in post-processing as a
result. I currently run with level 0 sharpening, but am finding
certain scenes have significant haloing which I would like to have
greater control over.

---------------------------------------------------------------------------------------

Hi Simon,

My non-technical tests on this topic have shown to me that in-camera sharpness settings below zero cause unrecoverable loss of very fine detail, at least to my skills in sharpening.

Perhaps someone else has a better sharpening technique than what I use, but as an example, the sharpened sections in file below (very small 100% sections of the original) were sharpened by applying USM to the Lightness channel in LAB mode which is supposed to be one of the best sharpening methods.

This image shows that the very fine detail (i.e. the chain links in the distant fence) in the '0' files has been lost in the '-2' file and I know of no way to pull it out.

If someone can do this I'd LOVE to know how.

Gordon
--

Charles C Lloyd · Oct 22, 2002

Gordon W said:
This image shows that the very fine detail (i.e. the chain links in
the distant fence) in the '0' files has been lost in the '-2' file
and I know of no way to pull it out.

If someone can do this I'd LOVE to know how.

Me too.

Looks like you may have debunked my testing by using a real-world picture. However, I will say I think the foreground elements (the green plastic chair) of the -2 case look better in your final result as there appears to be less noise in the plastic.

I don't mean to disparage your experimental techniques, but since you didn't mention a few things I am searching for a possible alternate explanation to the loss of detail. So, a couple of questions:

did you use focus lock before taking your series of shots? In my test, I used autofocus and then switched to manual so all shots would use exact same focus. I can imagine the autofocus mechanism having a bit of hysterisis between shots.

how sturdy is your tripod and was the wind blowing at all? I found in some tests I was doing yesterday that extremely minute camera shake was happening at small aperture settings (slower shutter speeds) when the wind would blow and vibrate my camera (I need to upgrade my tripod for the heavy 707).

related question: did you use the self-timer to trigger the shutter?

Thanks.

Charles.

Gordon W said:

------
The other day I went to... no wait, that was someone else.
Sony 707 (Since Oct 14, 2002)
Canon S100 (Since Feb 2001)

Charles C Lloyd · Oct 22, 2002

Gordon W said:
My non-technical tests on this topic have shown to me that
in-camera sharpness settings below zero cause unrecoverable loss of
very fine detail, at least to my skills in sharpening.

Gordon,

One other thng occurred to me that may back up your findings. I presume you used the fine jpg setting for storing the shot. Since the in-camera sharpening is probably done prior to jpeg compression, that could explain why the 0 setting gave better detail and -2 lost those details in the compression. Assuming no other abberations in the test (as mentioned in my previous message), I suspect that using tiff mode would have yielded different results (ie no loss of detail in the distant chain-link fence).

[BTW, I did my captures with tiff enabled, but I discarded the tiff images since the difference was small enough that it didn't warrant mention. However, since I didn't have any fine detail in my shots, I was probably fooling myself. Also, since the use of tiff is so unweildy, I figured it was a bit academic. Perhaps for studio work it makes sense to use tiff on the 707, but not in the real world.]

Charles.
------
The other day I went to... no wait, that was someone else.
Sony 707 (Since Oct 14, 2002)
Canon S100 (Since Feb 2001)

Gordon W · Oct 22, 2002

Charles C Lloyd said:
Looks like you may have debunked my testing by using a real-world
picture. However, I will say I think the foreground elements (the
green plastic chair) of the -2 case look better in your final
result as there appears to be less noise in the plastic.

Hi Charles, you bring up some good points here. For instance, if one is more interested in reducing image noise than in retaining fine detail, one might be ahead to use a softer sharpness setting and rely on heavier sharpening in post-processing. Hmmm...I'm going to have to look further into that aspect of this.

Charles C Lloyd said:
I don't mean to disparage your experimental techniques, but since
you didn't mention a few things I am searching for a possible
alternate explanation to the loss of detail. So, a couple of
questions:

Oh, my techniques have disparaged before, so nothing new. This was primarily a test that I did a couple weeks ago merely to confirm findings that Shay had come up with already so I didn't see much point in wasting a lot of time at it. I expect a search of the forum would turn up his more thorough results.

It was also a test to see if any of the various formats (Standard, Fine, TIFF) held more detail than the others because I thought I noticed in one set of files that there was less of the fine detail retained in Standard, but this did not end up being the case.

Charles C Lloyd said:
did you use focus lock before taking your series of shots?

I don't believe I did, although as I recall I half pushed the shutter release for each shot to ensure it had plenty of time to lock on target. This is my normal SOP, but you raise a good point. I will have to do this again with locked focus. I might have time tomorrow.

Charles C Lloyd said:
In my test, I used autofocus and then switched to manual so all shots
would use exact same focus.

One reason this never occurred to me is the distances involved and also because the camera has never missed a focus. I wish the EXIF data gave the focus distance, but I don't see it there.

Charles C Lloyd said:
I can imagine the autofocus mechanism
having a bit of hysterisis between shots.

Oh, I hope not. The camera had to be focused at infinity and I can't imagine enough hysteresis in a virtually new camera to effect it all that much. But it does present a potential variable that I should have thought to remove. It appears I thought of everything else.

Charles C Lloyd said:
how sturdy is your tripod and was the wind blowing at all?

Sturdy enough to have held a 4x5 studio camera for many years and from the position of the leaves in others of these shots I can say there was no wind to speak of plus these were taken from a fairly sheltered corner of my patio.

FWIW, the EXIF data says these were shot at full wide angle, manual mode at f4, 1/200, ISO 100.

Charles C Lloyd said:
I found in some tests I was doing yesterday that extremely minute
camera shake was happening at small aperture settings (slower
shutter speeds) when the wind would blow and vibrate my camera (I
need to upgrade my tripod for the heavy 707).

This would not have been the case with these shots.

Charles C Lloyd said:
related question: did you use the self-timer to trigger the shutter?

Yes. More of my SOP is to do this whenever I use the tripod. I also invariably shoot brackets if conditions allow and using the self-timer ensures each frame with line up identically if I chose to stack them for noise reduction.

So, it looks like the focus lock might be the weak leak in this test. I will try this again with locked focus although I fully expect the same results. However, when testing nothing should be left to doubt. I just wish I'd thought of the focus lock the first time.

Gordon
--

netguru717 · Oct 22, 2002

-2 is free of sharpening, all other do sharp. unfortunately also the soft differences which increases the visible noise and other things.

I always uses -1 with 707 and I am sure this is the best if you want not processing after and want to have one setting for all photos.

Software on PC is better but mostly not worth the work. Its more a subject of religion that your result here really is better of you add the pros and cons. So set -1 and do nothing more. Use the time to make photos.

Simon Glynn said:
Hi All,

SUMMARY OF THIS POST
------------------------------------------------------
I discuss in-camera sharpening on the 707 and wonder whether level
-2 sharpening is actually the least sharpened image and whether you
should move toward performing sharpening in post-processing as a
result. I currently run with level 0 sharpening, but am finding
certain scenes have significant haloing which I would like to have
greater control over.

---------------------------------------------------------------------------------------

The following two articles.

http://www.robgalbraith.com/diginews/2002-10/2002_10_20_eos1ds.html

http://www.microsoft.com/windowsxp/digitalphotography/learnmore/fixit/sharpening.asp

are in realtion to, or have a section relating to in-camera
sharpening. They identify that the inclusion of an optical low pass
filter has the effect of reducing undesirable digital artifacts in
images, but has the disadvantage of reducing the sharpness of the
final image. They indicate that an optical low pass filter is
likely to be present in most digital cameras so I'll make the
assumption that the 707 has this filter, based on its ability to
alter image sharpness in-camera.

The primary message in both of the articles, was that in-camera
sharpening is either not as powerful or not as flexible as
post-processing sharpness, and that leads to my question.

QUESTION: Is there a reason that the 707 would have the ability to
reduce sharpness of an image, beyond that which is natively
produced at the CCD (ignoring the RAW to TIFF/JPG conversion)? In
other words Is there reason to believe that -2 sharpening is a
deliberate bluring of the image, or alternatively is it the least
sharpened version of the captured image

I understand from previous posts, that there are those who find 0
level sharpening to produce an image that appears to be most
representative of the captured scene, however my own experience on
high contrast subjects generates photos with significant "halo"
sharpening artifacts when using 0 level sharpening.

I will hypothesise the following, and would be keen to hear
feedback in relation to the following and my question..........

HYPOTHESIS

If in-camera sharpening operates on the RAW CCD data (i.e. on the

5 million pixels, most of which are Green, some Red and some Blue)
then it may not be possible to match in-camera sharpening via post
processing (Why?, well we don't have access to the RAW data and
hence can't replicate or improve on the in-camera processing prior
to the Rpixel Gpixel Bpixel data being converted into an RGB per
pixel image).

If in-camera sharpening occurs AFTER the RAW -> combined image

conversion stage, then if minimal in-camera sharpening is applied,
it should be possible to recreate, or improve on in-camera sharping
via post processing (Why? in-camera sharpening has no adjustable
parameters, e.g. radius, threshhold, amount, and may be restricted
based on camera CPU processing power)

(I'm hoping this is the case) -2 in-camera sharpening is the

least sharpened image. It's blurred appearance is due to the low
pass filter in the optical path, and it should be possible to
achieve or exceed the sharpness achieved using 0 level in-camera
sharpening. ..... If I am incorrect and -2 is actually a BLUR
filter applied to the image, then it will not be possible to
achieve the same sharpness as level 0, as image data will have been
lost

Thoughts, and experimental results welcomed. I will try and find
appropriate subjects to perform the testing on myself.

Regards
Simon Glynn

Gordon W · Oct 22, 2002

Hi again Charles,

Charles C Lloyd said:
One other thng occurred to me that may back up your findings. I
presume you used the fine jpg setting for storing the shot.

Actually, the images I posted in this thread were from TIFFs.

Charles C Lloyd said:
Since the in-camera sharpening is probably done prior to jpeg
compression, that could explain why the 0 setting gave better
detail and -2 lost those details in the compression.

I hate to be dense but you lost me on that one. If this was true then there'd be no loss of detail in the softer settings in TIFF and this is not what I found.

Charles C Lloyd said:
Assuming no
other abberations in the test (as mentioned in my previous
message), I suspect that using tiff mode would have yielded
different results (ie no loss of detail in the distant chain-link
fence).

When I shot this series I shot at all sharpness settings in each of the three modes (Standard, Fine, TIFF) and as I recall there was no difference with the exception of the expected JPEG artifacts.

However, since you raise this point, I'll pull this series out of the archive and confirm this statement.

Charles C Lloyd said:
[BTW, I did my captures with tiff enabled, but I discarded the tiff
images since the difference was small enough that it didn't warrant
mention. However, since I didn't have any fine detail in my shots,
I was probably fooling myself. Also, since the use of tiff is so
unweildy, I figured it was a bit academic. Perhaps for studio work
it makes sense to use tiff on the 707, but not in the real world.]

Well, there is some small improvement in the TIFF images over the JPEG but it's got to be a pretty special shot before I'll bother. Another of my tests a couple months ago was comparing Standard, Fine, and TIFF and I got much better images from stacking 3 Standard shots than I got from one TIFF. If the subject is such that stacking is not feasible (i.e. the subject is moving) then TIFF would be the best choice.

Gordon
--

Gordon W · Oct 23, 2002

Charles

I ran a new test today and came up with basically the same results with the focus manually locked at infinity that I did in my previous test.

Actually, it looks too like what I see in these new images refutes what I've read in other posts on this subject in the forum so rather than post it here where it's more likely to be overlooked, I think this is important enough to warrant its own thread.

So if you're interested in seeing what I came up with, go to...

http://www.dpreview.com/forums/read.asp?forum=1009&message=3628563

Gordon
--

Simon Glynn · Oct 23, 2002

I worked on a couple of examples for an hour or so but it was inconclusive. In some sections of an image shot at -2, I was able to pull out the same level of detail as the 0 image, but in other sections of the same image the technique did not seem to work as well. I wasn't using inbuilt photoshop USM, but a technique described here which is a little fiddly

http://www.astropix.com/HTML/J_DIGIT/USM.HTM

Sorry that it doesn't add a lot to the discussion. Sony sharpening does appear to be quite clean and hard to replicate.

Regards
Simon

Simon Glynn · Oct 24, 2002

SUMMARY--------------------------------

Through an analysis of hot pixels in black frame images, I think I can show that level -2 is the original image captured at the CCD, that it is not deliberately blurred and that as a result it contains the most image detail. It therefore follows that it should be your image of choice to use as an original, but only if you are a bit of a star with an sharpening in your image editing program, as it isn't particularly easy to get it to match even that produced by the 707 on its level 0 sharpening mode. This is a litle loose, but all I can come up with at the moment.
---------------------------------------

In maths there are is a way of proving a theroem wherby instead of trying to prove that it is true, you instead focus proving that it can't be false. It then follows that the theorm must be true. Since I can't provide a mathematical proof that is watertight, mostly because I don't have the knowledge, I will offer a conjecture instead.

The conjecture is is that Level -2 sharpening is the least sharpened version of an image captured by the 707

There are a couple of things that I am unable to do

1. Determine how to show that a level -2 image can be converted into a level 0 image via post processing sharpening.
2. Determine the type of sharpening algorithm used to create final Sony images

Instead I have focussed on trying to show that
1. a level -2 image is a NOT blurred version of the captured camera image
2. no detail is lost in the captured image.

If the above can be shown, then it follows that via some means (post processing) it must be possible to convert the level -2 image into a level 0 image. How you do that is irrelevant to the conjecture.

To know that image detail hadn't been lost, I would have to understand exactly what image was recorded at the camera CCD. There is a way to do that. Take a picture of nothing.

"But if nothing is there, what good does that do?" you ask

What gets recorded to the memory stick is not the light that strikes the CCD, but instead is a numerical representation of the light that hit the CCD, and two things come into play in generating that numerical representation of the light.

1. Thermal Noise: Random noise caused by heat which affects the recorded value

2. Hot, Stuck, or Blown pixels. Elements on the CCD that are malfunctioning and always generating an incorrect value

You might see where I am going with this......

I found a hot pixel in my images, always appearing at position 940x , 1196y in the final image. I utilised a program called deadpixeltest to locate the hot pixel and to analyse the pixels around it.

QUESTION 1. Is a level -2 a blurred image of what is captured at the CCD?

Well if it is, then my hot pixel will not be a single pixel on its own, but rather a hot pixel, surrounded by an average of the hot pixel value and the surrounding pixels. However when analysing the surrounding pixels, I found them to be no different (on average), than any other pixel in the rest of the image. This indicated that the recorded hot pixel in the final image, had no blurring applied to it after being output from the CCD.

Now if this is true, then it must also be true for all other captured information at the CCD (i.e. image data).

I think that this points to Question 2 being answered, i.e. no image detail is lost in the recording of a level -2 sharpened image.

Thats basically it... The pixels can't lie ( I think

Level -2 sharpening contains the most original image information possible. How you get it to reflect a level 0 image is anyone's guess.

FYI, the level 0 image, when analysed the same way, also had the hot pixel in the same position, but surrounding it (above, below, left and right) were four pixels which had values 4 to 6 times the average of the remainder of the image. This makes sense as sharpening is not about simply adjusting contrast, but also about emphasising edges and highlights. One way to emphasise a highlight (the hot pixel) is to surround it with a layer of similar pixels and in extreme cases with high contrast pixels (i.e. opposite colours, resulting in halos). The fact that these pixels appeared at the east, west, north and south posistions, indicated that the algorithm worked only on the nearest neighbour pixels, and not the complete set of surrounding pixels (N,E,S,W,NW,SW,SE,NE)

Technical details.

Images captured
lens cap on
ISO 100
F 8
S 1000
Image TIFF
Ambient Temp 20-25C

Total of 7 images captured in each setting of level -2 and level 0 sharpening. The 7 images were layered on top of each other in photoshop and using the graduated stacking technique (100%, 50%, 33%, 25%, 20%, 17%, 14%) were reduced to single layers to reduce random thermal noise in the captured image.

Images then passed through deadpixeltest (Michael Salzlechner) and hot pixels were located in the same position in both images. Surrounding pixels in a 9x9 matrix were analysed to answer the questions in the above conjecture.

Regards
Simon Glynn

technolawyer · Oct 24, 2002

Simon: great work! Your methodology and proof seem reasonable. I would like to see you perform the same test on a (+2) sharpened black image to see if the sharpening halos are more pronounced than the (0) setting image. BTW, about how many dead pixels did you encounter in your image.

Kurt

Simon Glynn said:
SUMMARY--------------------------------
Through an analysis of hot pixels in black frame images, I think I
can show that level -2 is the original image captured at the CCD,
that it is not deliberately blurred and that as a result it
contains the most image detail. It therefore follows that it should
be your image of choice to use as an original, but only if you are
a bit of a star with an sharpening in your image editing program,
as it isn't particularly easy to get it to match even that produced
by the 707 on its level 0 sharpening mode. This is a litle loose,
but all I can come up with at the moment.
---------------------------------------

In maths there are is a way of proving a theroem wherby instead of
trying to prove that it is true, you instead focus proving that it
can't be false. It then follows that the theorm must be true. Since
I can't provide a mathematical proof that is watertight, mostly
because I don't have the knowledge, I will offer a conjecture
instead.

The conjecture is is that Level -2 sharpening is the least
sharpened version of an image captured by the 707

There are a couple of things that I am unable to do
1. Determine how to show that a level -2 image can be converted
into a level 0 image via post processing sharpening.
2. Determine the type of sharpening algorithm used to create final
Sony images

Instead I have focussed on trying to show that
1. a level -2 image is a NOT blurred version of the captured camera
image
2. no detail is lost in the captured image.

If the above can be shown, then it follows that via some means
(post processing) it must be possible to convert the level -2 image
into a level 0 image. How you do that is irrelevant to the
conjecture.

To know that image detail hadn't been lost, I would have to
understand exactly what image was recorded at the camera CCD. There
is a way to do that. Take a picture of nothing.

"But if nothing is there, what good does that do?" you ask

What gets recorded to the memory stick is not the light that
strikes the CCD, but instead is a numerical representation of the
light that hit the CCD, and two things come into play in generating
that numerical representation of the light.

1. Thermal Noise: Random noise caused by heat which affects the
recorded value
2. Hot, Stuck, or Blown pixels. Elements on the CCD that are
malfunctioning and always generating an incorrect value

You might see where I am going with this......

I found a hot pixel in my images, always appearing at position 940x
, 1196y in the final image. I utilised a program called
deadpixeltest to locate the hot pixel and to analyse the pixels
around it.

QUESTION 1. Is a level -2 a blurred image of what is captured at
the CCD?

Well if it is, then my hot pixel will not be a single pixel on its
own, but rather a hot pixel, surrounded by an average of the hot
pixel value and the surrounding pixels. However when analysing the
surrounding pixels, I found them to be no different (on average),
than any other pixel in the rest of the image. This indicated that
the recorded hot pixel in the final image, had no blurring applied
to it after being output from the CCD.

Now if this is true, then it must also be true for all other
captured information at the CCD (i.e. image data).

I think that this points to Question 2 being answered, i.e. no
image detail is lost in the recording of a level -2 sharpened image.

Thats basically it... The pixels can't lie ( I think Level -2
sharpening contains the most original image information possible.
How you get it to reflect a level 0 image is anyone's guess.

FYI, the level 0 image, when analysed the same way, also had the
hot pixel in the same position, but surrounding it (above, below,
left and right) were four pixels which had values 4 to 6 times the
average of the remainder of the image. This makes sense as
sharpening is not about simply adjusting contrast, but also about
emphasising edges and highlights. One way to emphasise a highlight
(the hot pixel) is to surround it with a layer of similar pixels
and in extreme cases with high contrast pixels (i.e. opposite
colours, resulting in halos). The fact that these pixels appeared
at the east, west, north and south posistions, indicated that the
algorithm worked only on the nearest neighbour pixels, and not the
complete set of surrounding pixels (N,E,S,W,NW,SW,SE,NE)

Technical details.

Images captured
lens cap on
ISO 100
F 8
S 1000
Image TIFF
Ambient Temp 20-25C

Total of 7 images captured in each setting of level -2 and level 0
sharpening. The 7 images were layered on top of each other in
photoshop and using the graduated stacking technique (100%, 50%,
33%, 25%, 20%, 17%, 14%) were reduced to single layers to reduce
random thermal noise in the captured image.

Images then passed through deadpixeltest (Michael Salzlechner) and
hot pixels were located in the same position in both images.
Surrounding pixels in a 9x9 matrix were analysed to answer the
questions in the above conjecture.

Regards
Simon Glynn

--
Kurt

Howdy · Oct 24, 2002

Simon Glynn said:
Well if it is, then my hot pixel will not be a single pixel on its
own, but rather a hot pixel, surrounded by an average of the hot
pixel value and the surrounding pixels.

Thats basically it... The pixels can't lie ( I think

Nice start, though I would suggest some comments. As I understand, your premise is that a 'hot pixel' will only affect itself. Therefore, if surrounding pixels are not affected, then you are 'seeing' what the ccd captured.

Several concerns. First, I'm assuming that you worked with a TIFF file. Otherwise, as good as fine JPEG is, it is still a lossy algorithm. While it will allow hot pixels to be identified, it also allows the information from the hot pixel to bleed over into surrounding pixels, and vice versa. (ie, source of JPEG halos.0

Second, the 707 is using a Bayer ccd. For each 'pixel' in your image, you are seeing a manipulation of the light captured by the individual pixel sensor on the ccd, and that manipulation looks to the surrounding pixels for information. And the opposite is true. The dead pixel affects the manipulation of surrounding pixels. In theory, all dead pixels should be mapped, and 'removed' from the manipulations, but that is theory. So, even at its most basic level, a TIFF image is not a 'perfect' recording of the light received at each sensor pixel.

So, I would say, the pixels 'can' lie. At least, they tell a manipulated verson of the truth.

Third, that said, your method is probably as close as we will ever come to 'mapping' the detail captured by a sensor pixel.

Thanks for your effort.

Howdy.

Simon Glynn · Oct 24, 2002

Yes you're right. I knew that each recorded pixel was a combined result of looking at R, G and B pixels in a specified pattern, but I don't know how they combine the pixels so couldn't work backwards to understand which pixel on the array might have been affected. The hot pixel was recorded as a grey value, so whatever process was used to combine the actual pixels to recorded pixels, it didn't retain the colour information of the hot pixel. This and the fact that no surrounding pixels appeared to have a "bleed" effect from the hot pixel (which would be expected in a basic combination of physical pixels, as you pointed out) points perhaps towards some smarts in the combination of pixels, disallowing the bleed.

And for your other questions ,yes I did use TIFF.

Howdy said:
Simon Glynn said:

Well if it is, then my hot pixel will not be a single pixel on its
own, but rather a hot pixel, surrounded by an average of the hot
pixel value and the surrounding pixels.

Thats basically it... The pixels can't lie ( I think

Click to expand...

Nice start, though I would suggest some comments. As I understand,
your premise is that a 'hot pixel' will only affect itself.

Simon Glynn · Oct 24, 2002

Kurt,

I appear to have only one hot pixel when capturing images at 1/1000 sec. At slower shutter speeds and higher temperatures I might have more than one . Haven't really checked to be honest. I might try the +2 sharpening experiment one day, though I would be far more interested to see if anyone can match the sharpening that the Sony produces, using a level -2 as a basis. My hot pixel had a luminance of around 50-60 (of a max of 255) so it isn't all that noticable.

Regards
Simon

technolawyer said:
Simon: great work! Your methodology and proof seem reasonable. I
would like to see you perform the same test on a (+2) sharpened
black image to see if the sharpening halos are more pronounced than
the (0) setting image. BTW, about how many dead pixels did you
encounter in your image.

Kurt

707 in-camera Sharpening - Low Pass Filters

Active member

Senior Member

Active member

Senior Member

Leading Member

Senior Member

Senior Member

Senior Member

Senior Member

Senior Member

Senior Member

Leading Member

Senior Member

Senior Member

Active member

Active member

Well-known member

Leading Member

Active member

Active member

Keyboard shortcuts