Recent epiphany in EXR blog article

Started Apr 17, 2012 | Discussions thread
Timur Born
Senior MemberPosts: 3,772
Like?
Re: sheesh ...
In reply to Kim Letkeman, Apr 20, 2012

Kim Letkeman wrote:

a) different analog gain/amplification settings on half the sensor (what you call ISO DR afaik) as opposed to

Benefit: Only the highlight half of the sensor is underexposed two stops. And it has the room to avoid too much noise

How does increasing analog gain before AD conversion avoid noise? Does that mean that at lower (especially base) ISO the analog gain is too low to convert noise free shadows? Does that mean that with exactly the same exposure time and aperture you get less noise in the shadows by applying more gain? And most important, how practically relevant is the difference?

or loss of too much data.

Let's discuss what should be the practical result of such an arrangement.

1) Loss of shadow data should get worse in the following order: L + DR 200/400 (full sensor underexposed), M + DR 200/400 (half sensor underexposed and binned), L + DR 100 (full sensor properly exposed), M + DR 100 (full sensor properly exposed and binned).

Can you demonstrate in images how this is the case?

Note: At any ISO above whatever the base ISO is for this sensor -- and we do not actually have that info, so this discussion is necessarily vague -- there is amplification applied to the sensor.

Amplification likely is always applied, even at base ISO, in order to meet the AD converter(s)' input characteristics.

All sensors are better with analog amplification to a certain point, then digital amplification is perfectly adequate. For the X10 and X100 that number is around 1000iso IIRC. So there is benefit to allowing the mid and low tones to be amplified in the analog domain.

One of the following images was shot at ISO 400 + DR 400, the other two were shot at ISO 100 + DR 100 and then digitally lifted afterwards. Where is the fundamental difference that would justify implementing more expensive and complex multi-amplification logic?

In the following thread I compared ISO 800 + DR 400 to ISO 200 + DR 100 after digitally lifting the ISO 200 JPG . The digitally lifted ISO 200 JPG still came out better than the untouched DR 400 one of which you claim "better analog than digital" performance.

http://forums.dpreview.com/forums/readflat.asp?forum=1012&message=40693141&changemode=1

b) running the whole sensor at lower gain/amplification (what you call software DR)?

I assume that you are aware that RAW data is linear, and is thus stored in stops that each take half of the amount of available data.

Yes, I am aware of that. There are non-linear design out there, though, like the NEX-7 is reported to be.

My point is that Fuji can allow the analog amplification to do its job so that the RAW data is stored in the correct stops. Blending can be thus a trivial algorithm (as it is with shutter based DR extension)

If the same algorithms were used then the output image would look more similar, which it does not. Exposure time DR images are visibly different from amplification based DR images in both noise and highlight gradations.

Neither the implementation of differently working amplifiers + converters nor the automatic blending are trivial. The camera itself fails to do the blending properly (leaving blooming clipped highlights of the longer exposed half intact, attenuated but still overwriting information) and the different raw converter programs have their own issues with doing it right.

instead of a very complex algorithm that requires the data to be spread out to fill the stops digitally before the blend can actually take place.

When the whole sensor is run at lower gain/amplification/ISO then all you have to do is take 12 bit raw data with 3500 - 4000 luminance gradations and map that down to an 8 bit JPG with 256 gradations after applying curves. Happens every day, happens in the non EXR X100, is the definitively much more trivial task.

Were I to specify such a feature, I would ask for each pixel's amplification to also be changed independently (ganged together by sensor half of course.). The feature would therefore provide the ability to get a lesser exposure on half the sensor in two separate ways. The up front additional cost is what? Somewhere between zip and squat.

So in your books implementation of an electronic shutter (turning off photo sites) allows to add additional and separate amplification/voltage logic (+ maybe corresponding extra AD converter logic) at no additional cost?! Maybe someone with a background in sensor/converter circuitry design can chime in? But I fear that most of these people are found in other sub-forums, not here.

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