D70 linearity

Here's one for the techies: does anyone know the linearity of response

of the D70's CCD? Also, the linearity of the compressed NEF file? By

linearity I mean that the signal out (of each pixel; signal recorded

in the file) is a linear function of the power incident on that pixel,

at least away from the extrema. Let's keep it at a fixed gain

setting, say ISO 200. Actually, I'd settle for a known function,

doesn't have to be linear. I'm using dcraw to convert to ppm, if that

matters. <a href=www.majid.info/mylos/weblog/2004/05/02-1.html>Fazal

Majid</a> provides decoding curves taken (I'm guessing -- haven't

checked) from dcraw.c, but labels neither abscissa nor ordinate, and I

don't know what they're showing, except that something somewhere is

actually not linear. Presumably the chip maker's SDK would have that

information, but so far the best I can figure is that the D70 uses a

Sanyo chip based on a Sony design for the D100(??), which doesn't help

much.

<p>

I know that the compression scheme is not lossless -- I'm not

interested in that (?) You could argue that the real quantity of

interest is neither the linearity of the chip nor the compression

scheme but the linearity of the convolution, and the convolution is

what you get and that you really should calibrate against known

values. I agree completely, but also figure that the more I know the

easier this gets. (The application is non-photographic, but the

instrument -- a D70 -- is, hence the question here.) Also, are there

any non-deBayerizing raw converters out there? Thanks for any info!

Dave, what the hell are you talking about? :>}

" but labels neither abscissa nor ordinate, and I don't know what they're showing, except that something somewhere is actually not linear."

 

Get it?

 

Errol

I'd expect logarithmic rather than linear response from the sensor to variations in luminosity.

 

However, I just take pictures. I'm not trying to modify a D70 into a phaser or anything.

 

Be well,

it used to be so simple. :>{

It's not difficult to measure the response curve yourself -- vary the exposure (by placing

different ND filters in front of the lens, for example) and look at the corresponding output.

Then you'll know the curve of your particular D70, not just some generic or

"average" curve. If you're interested in sensor (really A/D converter) output, you can

modify or configure dcraw to just read the raw numbers without doing a gamma

correction, etc.

 

Compression has no effect on the shape of the response curve (it better not!). It only

creates gaps in them.

The output non-linearity of a CCD is typically no more than 3% (it's usually better than 1%) when the signal levels are not near the limits of the sensor. Non-linearity is usually measured between 20% and 80% of the maximum signal level that the sensor can handle. This measurement is described in full details in many textbooks, but there's a brief <a href="http://www.ccd.com/ccd105.html" target="right">description of the test on Apogee's website</a>. The different signal levels can be simply obtained by changing the f/stops or shutter speeds. The gain test can get pretty tricky though...

<i>It's not difficult to measure the response curve yourself...</i> Well, you're right of course, and I will calibrate. But I'd still like to know just what's going on at each step. The concern is that I might not have sufficient dynamic range in the calibration I have, and knowing the behavior would help.

<p>

<i>Compression has no effect on the shape of the response curve.</i> Right again, I meant file format instead of compression scheme. Well, "format" isn't quite right either: rather whether a raw file has uncorrected data straight from the ADC (which is itself a digitized approximation of an analog signal, of course) or whether corrections are applied (which is what the "decoding curves" imply). The term "raw" seems to be a bit fungible.

<p>

Jim: Actually, usually I just take pictures with the thing, too. But sometimes you just can't resist using it for, um, nonstandard things, you know? Now, about that phaser ...

<p>

Gary: your first response might be the best yet! What the hell <i>am</i> I talking about?!

Sounds like you might be interested in trying some custom curves with your D70. Rather than trying to come up with custom curves yourself you can download curves created by other folks who are offering to share them free of charge.

 

If I'm recalling correctly you'll need to download the curve profile to your hard drive, then transfer it via USB cable to the camera with the menu set to the appropriate selection.

 

There's not a lot of really helpful info in Nikon's literature but other folks have worked hard to clarify how it works, what the advantages are and how to do it as easily as possible.

 

Here's one source:

 

http://www.planetneil.com/nikon/d2h-csm.html

 

Neil offers suggestions for curves that he believes are appropriate for the D70, D100, D2H and others. There are also links to other sites that discuss the issue of custom curves in greater depth.

 

At the very simplest a custom curve can allow you to bump the midtones during capture to your preference in order to avoid the usual post processing risk of blowing out highlights while trying to nudge the midtones.

 

Nikon Capture is required to make this work but a 30-day trial version of Capture was included with my new Nikon dSLRs.

Dave, Coffin's dcraw.c has a mode where it doesn't demosaic, and outputs a linear decompressed version of the D70 raw data. The compression curve is known, and it's in dcraw. It's also posted on an astronomy site that I could dig up for you, if you need a second opinion...

 

The little non-linearity the sensor has (as was mentioned, about 1%) is sensor dependent, and temperature dependent, you'll have to neasure it yourself. You'll find the shutter isn't 1% repeatable, nor is the aperture control mechanism, so your best bet is to use a good photometer to calibrat a light source that is pulsed in a time interval smaller than a target shutter time.

 

Have fun.

 

Joe

Hooking the D70 to a spectro?

Lex - this time... horizontal ;)

<i>Hooking the D70 to a spectro?</i> Not quite, but close: mapping 2D illumination fields. There are better, more expensive ways to do this, but I don't have to do it often. I thought Berg Na had a brilliant idea, using the shutter speed (I don't want to use aperture as that may change relative illumination, one more thing to calibrate), since that can range 4 orders of magnitude. Hadn't occurred to me as my other cameras are mechanical and kind of old: the main lens on the main shooter is 45 yrs old, tops out at 1/200 s, and with luck may get within 20% of that. You say not to expect better than 1% from the D70 -- well, that's OK, so long as it's not too much worse. I haven't used dcraw much -- I take it the "document" mode is the one that doesn't demosaicize? Thanks for the info!

<p>

Lex -- thanks for the curve info: might come in handy next time I take real pictures with the thing ...

Exactly. Document mode gives you 16 bit linear values, with no Bayer interpolation. So you'll get something that looks very much like the RGB pattern of the sensor.