«

Here is a question for anyone with a camera that has an ISO setting of
800 and greater. If you take an image at a low ISO setting (100 to
400) and use the levels adjustment to expand the histogram do you end
up with pretty much the same image you would get at ISO 800 or
greater?
And if you don't what are the differences?

jpc

On 3-Aug-2003, jpc <> wrote:

> Here is a question for anyone with a camera that has an ISO setting of
> 800 and greater. If you take an image at a low ISO setting (100 to
> 400) and use the levels adjustment to expand the histogram do you end
> up with pretty much the same image you would get at ISO 800 or
> greater?
> And if you don't what are the differences?

The problem is that a given image format only contains a fixed number of
values per color per pixel. When you adjust the exposure you change these
values. If you have an image that is underexposed, that fact alone may have
caused some of the dark areas to merge into black, when you try to lighten
the image you can lighten these areas but the detail has been lost and
cannot be recovered. Similarly, if you have an image with near white areas
and you lighten it further, the light areas will compress into white and
highlight detail will be lost. This is very similar to shooting negative
film. Exposure errors can be corrected to some extent in the printing
process, but the more extreme the error, the more you get muddy blacks or
blown out highlights.

Using 16 bit/color files helps, as do using raw files, because you get more
values per color. The best choice is to choose a sensitivity (ISO), shutter
speed and aperture that give you a properly exposed image to start with.

--
Tom Thackrey
www.creative-light.com

In message <>,
jpc <> wrote:

>Here is a question for anyone with a camera that has an ISO setting of
>800 and greater. If you take an image at a low ISO setting (100 to
>400) and use the levels adjustment to expand the histogram do you end
>up with pretty much the same image you would get at ISO 800 or
>greater?
>And if you don't what are the differences?

Let me rephrase your question. Are you asking what the difference is
between shooting a picture properly exposed for ISO 800, vs the same
shutter speed and aperture at ISO 100 with software correction of
levels?

The latter will have bigger steps between light levels, or higher
quantization. This will make more noise and confusion between image
detail and noise than if you shot at ISO 800 in the first place.

Higher ISO settings always result in more noise, but only if the images
in comparison are all properly exposed. A dark ISO 100 picture doesn't
have a lot of noise while it's dark, but the noise increases when you
bring the levels up, and you're always better off shooting at the
highest ISO that allows proper exposure.
--

<>>< ><<> ><<> <>>< ><<> <>>< <>>< ><<>
John P Sheehy <>
><<> <>>< <>>< ><<> <>>< ><<> ><<> <>><

In message <>,
jpc <> wrote:

>I understand and agree with all you are saying but let me expand on my
>question a bit.
>
>Lets say for some reason I'm forced to take a picture in very low
>light without a flash at ISO 100. When I examine the histogram it is
>1/8 the value it would be if I had enough light to expose the picture
>properly

ISO100 with the same amount of light does not result in levels that are
1/8th what they would be at ISO 800. They would be about 39% as high.

The levels are not proportional to the amount of light; the levels to
the 2.2 power are proportional to the light.

>(32 bit of histogram instead of 256 bits)

Those are not "bits". They are levels, and you'd have them up to about
99 in your scenario (ISO 800 exposure levels at ISO 100).

>Now lets say it is
>on top of a noise level of say 2 bits. So my S/N is 16.

>If I increase my ISO to 400 I'll up the amplifier gain from 1 to 4 and
>if I take another picture under the same conditions my histogram is
>now 128 bits and my noise is now 8 bits
>
>If I use the levels adjustment to expand the 128 bit histogram to 256
>bits I'll create a comb where there is a one bit gap between each real
>value and my colors or greyscale will be steppy. And obviously if I
>take a third picture at ISO800 I wouldn't have the pesky gaps in the
>histrogram. But since my noise is now 16 bits I don't think I would
>have gained that much more detail in the picture,
>
>Now if I use something like the photoshop noise filter to add one bit
>of noise to the adjusted ISO 400 picture so my color
>or grey scale translations are smoother I've increased my noise to 9
>bits and my S/N is now 15.
>
>While this is slightly worst than the S/N I'd would have gotten if my
>camera had a n ISO of 800--which it doesn't-- it's not that much
>worse.

Your logic, math and experiment are flawed. There's no point in
following your thought experiment, because it exits the real world at
step 1.

>So my question is, in the low light conditions where anyone is likely
>to use ISO 800 or greater, are the pictures you would take using the
>methods I've discribed all that much different?

You get the most noise-free results when you shoot at the highest ISO
that allows you to get a picture exposed the way you want it to be
exposed. A camera would have to be designed by a saboteur to give
better or even equal results by under-exposing at a lower ISO.
Under-exposing at the lower ISO quantizes both the signal and the noise
to a greater degree, creating even more noise, and making it difficult
for the brain (and software) to separate signal from noise.
--

<>>< ><<> ><<> <>>< ><<> <>>< <>>< ><<>
John P Sheehy <>
><<> <>>< <>>< ><<> <>>< ><<> ><<> <>><

On Tue, 05 Aug 2003 23:04:18 GMT, wrote:

>In message <>,
>jpc <> wrote:
>
>>Here is a question for anyone with a camera that has an ISO setting of
>>800 and greater. If you take an image at a low ISO setting (100 to
>>400) and use the levels adjustment to expand the histogram do you end
>>up with pretty much the same image you would get at ISO 800 or
>>greater?
>>And if you don't what are the differences?
>
>Let me rephrase your question. Are you asking what the difference is
>between shooting a picture properly exposed for ISO 800, vs the same
>shutter speed and aperture at ISO 100 with software correction of
>levels?
>
>The latter will have bigger steps between light levels, or higher
>quantization. This will make more noise and confusion between image
>detail and noise than if you shot at ISO 800 in the first place.
>
>Higher ISO settings always result in more noise, but only if the images
>in comparison are all properly exposed. A dark ISO 100 picture doesn't
>have a lot of noise while it's dark, but the noise increases when you
>bring the levels up, and you're always better off shooting at the
>highest ISO that allows proper exposure.


Thanks for giving my question some thought and coming back twice.

My logic, whcih may have some hole in it, is this;

1--I'm taking an image of a very low illumination scene which, to use
my original numbers, has a S/N of 16 at the CCD.

2--Because of the low S/N all I need is 32 bits of A/D to extract all
the useful information from the CCD. Any A/D resolution beyond that
only chops up the random noise into finer chunks.

3--If I take the image as it came out of the camera and use some post
processing proceedure--more complicated than a simple levels
adjustment--to add enough random noise to replace the random noise I
lost I will end up with essentially the same image I would have gotten
if I had turned up the ISO gain amplifier on the camera and digitized
the image using the full 256 bits of the A/D. In other words the
random noise I throw in afterwards is equivalent to the random noise
generated at the CCD.

{Am I confusing my nomenclature? An 8 bit A/D translates into 256
seperate reading and a 5 bit A/D translates into 32 seperate reading.
As I reread what I just wrote I think I should use some other wording
instead of "32 bits" and "256 bits" but for the life of me I can't
think what the proper wording should be. Must be too early in the
morning.}

Anyway, that is the simplified version of my thought experiment.

In the real world I'm trying to work out how my camera-- an Oly
3020Z-- works. From some previous experiments and from what I've been
told and read I have a CDD with a yet to be determined noise floor
(the electron noise in all its flavors) and yet to be determined well
depth. After the CCD there is a 1X to 4X linear amplifier (ISO 100 to
400) and a 12 bit A/D. Once the image has been digitatized the
camera's computer maps the 12 bit CCD output into the standard 8 bit
color color output in a more complex way than a simple gamma
correction. To add a bit more confusion this is controlled by the
contrast setting which has 11 different settings.

At least that is what I think is going on. If anyone knows better
shoot me down.

Once I develop an experimental method of assigning real numbers to
the camera I own now-- I've located some buggy but free image
analysis software on the net and am accumulating test targets-- I'm
going to pull down images from the review sites and see what I can
find out from them.

Finally, if everything works out, I'll lug my test targets down to the
my local camera store, and take some test images. Once I've analysed
them I decide what to buy to replace the 3020Z.

jpc

Have a look at a program called Neat Image

In message <>,
John T <> wrote:

>Have a look at a program called Neat Image

Neat Image might be good for removing some noise, but the discussion was
about _avoiding_ noise.
--

<>>< ><<> ><<> <>>< ><<> <>>< <>>< ><<>
John P Sheehy <>
><<> <>>< <>>< ><<> <>>< ><<> ><<> <>><