Unbiased noise test.

Started Feb 28, 2009 | Discussions thread
(unknown member) Contributing Member • Posts: 873

sidelight wrote:

Rcihardc wrote:

No not bias, but their measurements are just readings that are not
normalized and thus not suitable for comparisons, if you are a
chemist you should immediately recognize that measurement results
need interpretation before any conclusion can be drawn, you can't
just measure two values and say A is more than B. for example if you
have a elemental detection tool that can say measure 1 ppm -100ppm of
some atom concentration, if you see it show 1 ppm will say you have
1ppm? No because the true concentration is below 1 ppm and your tool
cannot read it. There are many issues like this with DXO, just go to
their website and read it...

I am not following you. They are comparative measurements using the
same test procedures. They say they are reproducible. If I setup a
measurement test, there would be a calibration curve and a limit of
detection established against which sample values would be compared
to determine results . I should think DxO does this and is able to
draw conclusions but if you dispute it, why not post the specific
flaws you see so people, including DxO, can address this?

The noise in the analog readout of a CMOS image sensor is different
from a continuous reading of a sensor in real time where there is
spikes of noise and interference. It is just a different world, you
cannot really compare directly, please read the Clark articles, there
is no threshold like that...

DxO only measure by looking at raw images, they can never measure
real noise which is only possible by disassembling the camera and
wire-bonding the sense amplifier lines to an Agilent spectrum

Again, I am not following you. I'm not a sensor or electronics guy,
but I find it difficult to understand when you say the sensor noise
cannot be measured using their procedure. I can understand that
there may be different ways to measure noise and under varied
conditions which they may not cover completely. Noise should
manifest itself in signal variation and have a standard deviation.
At low light conditions, I understand it to be mainly read noise. At
higher levels it is a mix of read and photon shot noise. Here is
DxO's noise statement:

you don't get it DxO is not measuring any signal, they are looking at a raw files, a raw file is not an analog electrical signal. Also shot noise power is proportional to sqrt(N) N being the number of photons detected, signal power is proportional to N, so S/N for shot noise is prop. to sqrt(N) which is high at good light and low at low light (that is why you can't have a single photon detector becuase you would never know when it arrives). So shot noise is important at low ligh not at high light. Clearly you did not read any of the Clark's article, please READ the article first and you will undrestand why DxO is not a good benchmark.

"The gray level output is often 0 or close to 0 for very low
illumination, but even if the sensor does not receive any light,
there still can be fluctuations in the measured signal (due mainly to
electronic and thermal factors) that can produce a high noise

Signal to noise does matter as does dynamic range. You can't see the
dynamic range of the D3X, 5DII, or the A900 on a monitor or in a
print. You would have to under and overexpose images and then show
what is recoverable. Or, you could measure this.

Don't the camera makers have sensor tests and quality control? They
don't just evaluate images. I bet Nikon and the others have design
goals supported by engineering feasibility studies, and they measure
everything to see that their goals are met. The sensor data, if done
with valid tests and well controlled comparisons, should be the
starting point because they are numeric, high precision tests, with
protocols that can be peer reviewed for relevancy. And they should
be reproducible.

Image evaluations are most certainly relevant, and as I said, I
ultimately want to see those as well, when controlled comparisons are
performed. Many of these type of tests are not well controlled
however or incomplete, and most user tests fall into this category.
The image evaluation attempted here, on the face of it, looked nice.
If there were not exposure issues, it would be a good comparison to
draw conclusions from. I would bet that image results would track
well with the DxO results, if test conditions are similar.

It should be possible for someone (users, not dpr) on this site to
compare two or more cameras (borrowed or rented if necessary) under
the same shooting conditions following an agreed upon test procedure.
I say should be, because somehow I think there would be endless
arguments about how the tests should be conducted. However, there
are other sites like diglloyd or LL for example, that do this.

-- hide signature --


-- hide signature --

Richard, NC
Never comment on something you don't know about

Post (hide subjects) Posted by
(unknown member)
(unknown member)
(unknown member)
(unknown member)
(unknown member)
(unknown member)
(unknown member)
(unknown member)
(unknown member)
(unknown member)
(unknown member)
(unknown member)
(unknown member)
(unknown member)
(unknown member)
(unknown member)
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow