Eric Stevens wrote:
On Sun, 31 May 2009 03:10:56 -0800, (Floyd L.
Davidson) wrote:
When are you going to mark your deletions.

You still haven't figured out how it works, have you.

What do you think I meant by 'statisticsl error limitations'?

You meant to imply that using big words with no meaning
will make it sound like you understand something you
don't. But anyone who does understand the process can
see that you don't.

Yet you claim it is possible to create an image from data that isn't
there!

I have never said any such thing, and see no point in you
making up distortions rather than discussion the topic at
hand.

You said the "source image" is what is projected on the
sensor. Interpolation of course is the method by which
the sensor data is converted to an image format for
viewing.

That's part of what I meant when I wrote of "the rules inherent in the
camera's software".

And you haven't yet figure out that one is the input to while
the other is an output from.

It is quite possible to change the raw data to effect
while balance, but it isn't normally done that way (Nikon,
for example, has hinted that they might be doing exactly
that in hardware).

... and therefore it is a different image. But nevertheless there is
only the one image which can be created from a set of unmodified raw
data.

But clearly that is not true. The raw data set does not
define one single image. It can be interpolated to
produce an image. But the interpolation can be done in
a nearly infinite number of ways, each of which produces
a *different* image. No one way is the _right_ way,
they are all just as correct as the next.

What is changed is the interpolation of the data when
creating an image format. The raw data is not changed,
and the raw file stays exactly the same. The way the
data is manipulated during interpolation changes.

An interpolated data set is a new data set.

The interpolation does not produce a new raw data set.
It produces an unique image.

But only the one JPEG can be created from the RAW data providing the
rules of the transformation do not change.

There is no one set of correct "rules of the
transformation".

False. Every different raw converter design uses a
different set of "rules". Coffin's dcraw.c uses one
set, Nikon uses another, and several other raw
converters are different from both of those.

But they are working on the camera's saved RAW file, not the
relationship between what the sensor sees and the saved RAW file.

Exactly. So why are you claiming otherwise? The raw
data set is not changed. But there are multiple,
correct, different sets of rules used to generate an
exact image from the raw data.

The sensor locations are hardly irrelevant either. As I
said, at least *nine* of them are used to generate each
pixel in the resulting image, and you can be assured the
location is relevant! It isn't one pixel and then 8
other randomly chosen locations... it's a group of 9
(or more).

So?

So please cease this silliness where you claim the
sensor locations are irrelevant.

The signals generated by the sensors are determined by the
rules inherent in the camera's software.

They are determined by rules inherent in the camera's
hardware. The sensor is not manipulated by software
other than clearing it and reading it. A given amount
of light on one sensor locations produces *exactly* the
same output from the sensor regardless of the camera's
software.

I should have said "The signals generated by the sensors are
-interpreted- by the rules inherent in the camera's software". To that
extent they are 'determined'.

I quoted you exactly above. Now you want to change what
you said.

Regardless, you are still wrong. The signals from the
sensor are interpreted according to *hardware* and the
resulting data set is written to a RAW file format.
That is what is "interpreted" by software.

As I have already said, there
is a one to one correspondence between the source image and the RAW
file.

You can say that all you like, but it still requires at
least *nine* different sensor locations to generate data
for each pixel of the resulting image. It is not a one
to one relationship.

I see the problem. You misunderstand what I mean by 'one to one'. By
that expression I mean that one imgage transforms into one RAW data
set. Its not as though the transformation entails (say) a quadratic
equation where the one image can give rise to either one of two RAW
data sets. See http://www.yourdictionary.com/one-to-one

So you now admit that it is not software at all, but a
hard wired hardware transform.

By next weekend we may force you into writing something
that is clear enough to make some sense.

You don't have a choice of RAW files for a given image. Nor do
you have a choice of images for a given RAW file.

But you have a choice of an infinite number of resulting
images when the camera raw data is interpolated. None
of them are exactly the same as your "source image" that
was projected onto the sensor.

... and none of them are the image defined by the RAW data. Close,
maybe, but not exact.

That is precisely what I've been trying to get through
your head! Good. Now you can get on with a sane
discusssion of raw data processing.

The raw data does not define one specific image. When
the data is interpolated there is then an image!

--
Floyd L. Davidson http://www.apaflo.com/floyd_davidson
Ukpeagvik (Barrow, Alaska)