Once in a year opportunity : ) .... messed up. : (

Started Dec 1, 2013 | Discussions thread
GeorgianBay1939 OP Senior Member • Posts: 3,890
Re: Once in a year opportunity : ) .... messed up. : (

texinwien wrote:

GeorgianBay1939 wrote:

clengman wrote:

I don't know if this

read noise decreases as ISO increases.

is so universal as to lead to the tactic of reducing read noise by increasing ISO until the highlights are clipped? In other words set the EV using artistic reasons (DoF, Motion Blur) and increase ISO until the rights side of the histogram clips.

It's not entirely universal. For instance, on an ISO invariant (aka ISOless) camera, read noise won't decrease as ISO increases. For many (most?) ISO variant (aka ISOful) cameras, read noise only decreases up to a certain ISO, after which it neither hurts nor helps you to raise ISO, as long as you don't clip important highlights.

Thank you.

I need to apply the above to my two cameras GX7 and GH2, both of which I want to learn how to set my ISO (in challenging circumstances, low light, large difference in scene luminance) to get optimal gain without blowing highlights. Assume that my EV setting looks after DoF and Motion Blur (which I'll call Artistic) variables. By optimal gain I mean the best possible to give me the lowest noise in the shadows when post processed (in Lightroom)

GH2 courtesy Sensorgen :

GX7 courtesy Jack Hogan :

Below I've tried to compare GH2/GX7 by closest measured ISO(Ssat) in terms of ReadNoisee- compared with DR (= Full Well Capacity/ ReadNoise)


Nom.ISO---------- MeasISO---------- ReadNoise e------------ DRstops

160/200---------- 167/153 -----------6.0/3.9 -------------------10.9/11.7

400/400---------- 413/308 -----------5.1/2.7 -------------------10.0/11.3

800/800---------- 817/611 -----------4.5/2.3 -------------------9.1/10.5

1600/1600------- 1538/1216 --------4.4/2.1 -------------------8.2/9.6

Hogan has gone one step further by finding the igain (apparently the number of e' required to produce 1 ADU), which appears to be different from Martinec's definition of gain: The conversion factor between the number of photons captured by the sensel and the raw value in ADU. If the number of photons captured is 1000, and the raw value is 100, the gain is 10 (1000 photons = 100 ADU x 10 photons/ADU) .... which is the inverse of what I normally think of as "gain".

process from sensel to data register after ADC.

If the process is: photons ->QE ->photoelectrons -> igain -> ADUs

Then photons -> (QE).(gain) -> ADUs [In other words Martinec definition goes directly from photons incident on the sensel to the ADUs.] IS THIS CORRECT?

Although interesting the above isn't getting me any further to develop tactics on how to set ISO when in challenging low light (high noise) situations. How much to lightening to do in post processing vs how much to do in the camera by increasing gain (ISO)????? (assuming that I am below the blowing/clipping highlight threshold)

For the E-M5, for instance, read noise has been measured to fall from ISO200 to ISO400, from ISO400 to ISO800 and from ISO800 to ISO1600, after which it seems to stay pretty constant.

or even increases (going to ISO3200) [I'll not look at measured, but I get the idea)

If I'm at ISO200, have my other settings 'right' and realize that I'm not going to blow any important highlights (live view blinkies for the win here), I can bump the ISO to 400 to get less read noise, as long as I'm still not clipping important highlights.

Ok, you are looking through the EVF at the LV histogram (Ignore blinkies for the moment, please) with your E-M5 on a tripod.

You are looking at a moving train, similar to the my nemesis .... moving bright lights against a black background, except for some snow (which you want in the final image).

You want to avoid blowing/clipping the lights but you want a bit of motion blur (which determines your shutter speed, say 1/30). You want as much light exposure as possible to illuminate the snow.

You are shooting in M(anual) mode so you don't care at all about what the camera is metering (since the meter is trying to average out that black frame to give you 18% gray in the output resulting in a huge overexposure.) You are smart enough to realize that once you get the motion down pat (1/30s) you have to decide on your lens and your ISO. You set your lens to f/4 because you've done this for the past 10 years and you know that will prevent you from oversaturating the highlights, while giving you maximum exposure (in EV) on the sensor for the luminance of the scene, without blowiing highlights. You stay at base ISO 200. At least that is the way it was "last year"!

Under the above conditions the histogram is jittering around just touching the right end of the box you take the shot. RIGHT? Same as last year. Wave to the train, pick up your tripod, go home, process the RAW file and send it to your friends on DPR!

However, (you are at base ISO) under the above conditions and the histogram is jittering around, well (say 2 stops) left of the right end of the box (the generator output for those lights has been turned down by the new president of the CPR to save fuel) you have three choices:

  1. crank up the ISO to "fill those shadows" using gain (ISO increased 2 stops to ISO 800 )
  2. open up the lens to "fill those shadows" using exposure (EV) by 2 stops to f/2
  3. leave it the way it is, shoot the %$#ing thing and open up the shadows in post processing.

My answer, I think is:

Use 2.) if possible subject to artistic constraints, then 1. and 3. depending on the ISOless-ness of the camera. If I were using the

E-M5 in going from 200 to 800 I would get RN 6.5->2.8 and DR 11.9->11.1

GH2 ...............................................................6.0->4.5 and DR 10.9->9.1

GX7 ...............................................................3.9->2.3 and DR 11.7->10.5

Compared to the E-M5, in going two stops from Base:

Both Pannys don't have the nearly the % drop in RNe-.

Both Pannys have a larger drop in DR.

I don't really understand how the RN works its way through the system in terms of shadow recovery.

I think that losing over a stop of DR means that the noise floor in the RawDigger histogram has moved up to the right one stop ... in other words a stop of data in the shadows has been lost.


This logic continues for ISOs 800 and 1600, after which increasing the ISO won't hurt or harm the image, as long as important highlights aren't clipped.


How does this relate to Dynamic Range?? Does that determine if I even have the option of the above --- if highlight clipping is critical to me? In other words is the luminance of the highlights so high that my artistic settings (EV) still will cause clipping, even at BASE ISO? In that case I would leave the setting at BASE and either shorten the shutter interval or close the f/ down. If I were on a tripod and WANTED some motion blur in the train lights, I would stop the f/down and set the ss to cause the right amount of blur.

If you are going to blow important highlights at base ISO, you'll have to stop down, choose a faster shutter speed, use an ND filter to cut some of the light or else resign yourself to living with the blown highlights.

Yes, ok.

This is where the idea of arbitrarily low ISOs (see also here) that Anders W is always talking about becomes very interesting. When this or a similar technology makes it into consumer digital cameras, the base ISO DR advantage of larger-sensor cameras will be a thing of the past.

I went to the Rambus Binary Pixel Imagers site to see what has happened since last February ....

Variable temporal oversampling takes multiple samples during a single exposure period to avoid pixel saturation improving the sensor's signal-to-noise ratio and low-light performance for better indoor and nighttime photography.

Seems to be a bit more than the re-setting concept that Anders W talked about last spring.

Very fast circuitry for short exposures, eh? But I suspect that the above feature would only make sense for low average scene luminance resulting in moderate exposures.

So the effective FWC would be several times bigger than current. Clever.

This approach at Digital Pixel Systems Technology seem to be a bit more general....

The core invention in the Digital Pixel System technology is the inclusion of an analog-to-digital converter (ADC) within each pixel of the image sensor. The ADC translates the light signal into a digital value at the immediate point of capture, thus minimizing signal degradation and cross-talk in the array and allowing for greater noise reduction methods. Once the data is captured in a digital format, a variety of digital signal processing techniques are used for optimal image reproduction.

Because each pixel has its own analog-to-digital converter and the information generated is captured and processed independently, each pixel in effect acts as its own camera. The exposure time for each pixel is adjusted to handle the unique lighting conditions at that pixel location in the image sensor array. A product built with the Digital Pixel System platform essentially has hundreds of thousands of individual cameras, each of which produces the best image possible. These images are then combined to create a high quality video frame or picture.

With this technology, you could just dial ISO down to 25 or 12.5 or however low you wanted to go to make sure you didn't blow any important highlights.

In the meantime can we have any hope that the expert photographic community will support

this sort of thing ?   Or at least give photogs the data right off of the sensor in LiveView.

I hope so.

Thanks a lot for your response and reading my musings above.  Am I on the right track?  I feel that I am deep water here!!!!


 GeorgianBay1939's gear list:GeorgianBay1939's gear list
Panasonic FZ1000 Panasonic Lumix DMC-GH2 Panasonic Lumix DMC-GX7 Panasonic Lumix DMC-GH4 Panasonic Lumix G Vario 7-14mm F4 ASPH +10 more
Post (hide subjects) Posted by
(unknown member)
(unknown member)
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow