ISOless sensors, read noise and photography - many questions!

Started 9 months ago | Discussions thread
Forum ProPosts: 28,111
Re: ISOless sensors, read noise and photography - many questions!
In reply to boardsy, 9 months ago

boardsy wrote:

bobn2 wrote:

There are other factors to take into account, in no particular order:

- on those graphs the 'base ISO' is not always too accurate, because the sensor may have been run into non-linearity.

- there are qualitative factors as well as the measure of read noise to take into account.

Sorry, this is all a bit opaque to me - what is the upshot for practical purposes?

Well, the real upshot is, you might use those graphs to help you select, if you value 'ISOless', but you're best off doing some test shots (as there is even sample to sample variation) to work out how best to use your camera). Things like pattern noise and pixel response non-uniformity are not included in the above graphs. Canon's particularly are bad for pattern noise, which means their useable range is less than indicated.

- is it possible or likely that the camera's ISO implementation is better tuned to its sensor and RAW data than your RAW software? Aside from highlight preservation on exposure, are there other benefits to "doing ISO" in post?

It's possible. Nikon, for instance, does some subtle gain adjustments with its cameras. Presumably they wouldn't be doing them unless the engineers thought that they brought some advantage. BTW, I think it's worthwhile distinguishing between 'processing' and 'post processing'. I think the brightness setting is best done in processing (raw conversion), not post-processing.

Do you mean: processing means RAW software (ACR, LR etc), and post-processing applies to JPG (PS, GIMP etc)?

Software packages mix and match. What I mean by 'processing' is the act of assigning brightnesses and colours to the pixel values, which are essentially photon counts. Basically, you want to assign the brightness you want at that stage, rather than later. Essentially, one might expect the processing results in a tiff or jpeg file, post processing would take that file and work with it. The boundaries are a bit confused because many processors (raw converters) include operations subsequent to conversion, and many PP tools can convert a raw image.

- for a given sensor may it be more effective to raise ISO to some minimum/optimal level vis-a-vis its read noise?

Yes, I think that's true. The D7000 above is better at 200 for instance.

But it has less DR & Sat Cap and equal read noise at 200 vs 100?

Sur, but saturation capacity and DR are only of interest if you're going to use them. If your scene and chosen exposure doesn't include anything that will blow out at 200, then 200 is as good as 100, and maybe a bit better in some cases.

- how do us non-scientists interpret these sensor graphs? E.g. for the Sony NEX C3 (probably the same sensor as my F3, maybe, lol), read noise actually drops down to ISO800 ...but so does DR! Is ISO400 the best compromise here, or should best DR (at ISO200) trump lower read noise (at ISO800)? And could this non-base ISO be valid for all shooting, or just low-light where optimal exposure becomes problematic?

DR trumps lower read noise if you need it, which is one of those judgements based on experience. A rule of thumb which covers most cameras pretty well is go with ISO but stop about 2 stops faster than base (3 for an APS-C Canon, 4 for a FF one).

You mean raise ISO in-camera as required, up to 2 (APS-C), 3 (APS-C Canon), or 4 (FF) stops?

the 4 stops for FF was Canon only. Yes, essentially use that as your fixed working ISO, lowere it if you need more highlight headroom.


- finally, could all this be solved for a given sensor by simply shooting a set of fixed aperture/shutter shots at each ISO level, normalising the resultant RAW brightnesses (with the, er, "exposure" slider :-|) and comparing noise (in the shadows, I guess?) by eye?

I think personal tests to one's own requirements are always the best - part of getting to know your camera. Just to add, the same goes for you raw processing tools.

Does what I outlined above make sense?

Yes, you'd be characterising your own workflow, which always makes sense. On the basis of what you learn, you might decide to change your workflow a bit.

Thanks for your help! One other point - why would ISO800 read noise be better than base ISO200, as in the NEX-C3 sensor above? Seems counter-intuitive - something to do with good ADC or other sensor gain up to that point, and just digital processing (brightness mapping) applied after that? Just vague ideas of mine, not an engineer!

The question is at which stage in the signal chain does the noise arise. Typically high low ISO read noise is caused by high late chain noise (say from the ADC). Raising the analog gain before that stage makes its noise less significant. The Sony column ADC's are very good, but ironically Sony seems to get less out of them than other manufacturers. Perhaps they run them faster, or are a bit less cute in optimising the gain settings.

-- hide signature --


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