Go Back   Steve's Digicams Forums > Digicam Help > General Discussion

Reply
 
Thread Tools Search this Thread
Old Feb 9, 2012, 9:38 AM   #1
Senior Member
 
MartinSykes's Avatar
 
Join Date: Feb 2010
Location: Cheshire, UK
Posts: 854
Default Idea - Dynamic ISO

I was wondering just how much of the flexibility of a digital camera is crippled by the need to align with old film standards. Is there any reason why for a given shutter/aperture the camera couldn't just count the photons and *then* scale to the 'perfect' ISO that would clip say .01% or some other user-definable amount of the image. So the ISO is determined after the shot based on the shot itself instead of before the shot based on the more approximate metering.

You could just replace the 14-bit 24MP sensor with a 28-bit 12MP sensor. Same amount of data but imagine the huge dynamic range it could handle and you'd never have a clipped highlight or shadow ever again.
__________________
AutoHDR - Free HDR software for Windows at http://www.autohdr.co.uk on Facebook at http://www.facebook.com/home.php?sk=...36045126467361
My Flickr stream: http://www.flickr.com/photos/martinsykes
Panasonic DMC-FZ38, Sony ­α580
MartinSykes is offline   Reply With Quote
Sponsored Links
Old Feb 9, 2012, 9:54 AM   #2
Senior Member
 
Join Date: Sep 2005
Posts: 1,093
Default

You do know that doubling the bits isn't at all comparable to halving the number of pixels, don't you? The usual rule of thumb is that halving the number of pixels would increase the dynamic range by a factor of 1.4, or less than ONE bit.
tclune is offline   Reply With Quote
Old Feb 9, 2012, 10:22 AM   #3
Senior Member
 
MartinSykes's Avatar
 
Join Date: Feb 2010
Location: Cheshire, UK
Posts: 854
Default

I think if you halved the pixels but remained with 14-bit processing then you're right - it doesn't make a huge difference but if you used the halving of pixel count to facilitate an increase to 28-bit processing at no net change to the amount of data per shot then that should give you the ability to represent *much* more range.

In simple terms, a pixel that previously became blown because it had reached the limit of 12 or 14-bits could keep counting until it reached the limit of 28 bits.
__________________
AutoHDR - Free HDR software for Windows at http://www.autohdr.co.uk on Facebook at http://www.facebook.com/home.php?sk=...36045126467361
My Flickr stream: http://www.flickr.com/photos/martinsykes
Panasonic DMC-FZ38, Sony ­α580
MartinSykes is offline   Reply With Quote
Old Feb 9, 2012, 12:49 PM   #4
Senior Member
 
iowa_jim's Avatar
 
Join Date: Jan 2010
Location: North Central Iowa
Posts: 589
Default

It sounds like that approach eliminates the need for a shutter- each pixel sensor gets it fill then stops. This is an intriguing idea. I wonder if the processing power of cameras is up to it, but if not yet, certainly someday. Haha, now HDR will be considered natural photography, refering to an earlier discussion on post processing vs 'real photography'.
iowa_jim is offline   Reply With Quote
Old Feb 9, 2012, 3:40 PM   #5
Senior Member
 
VTphotog's Avatar
 
Join Date: Mar 2005
Location: Extreme Northeastern Vermont, USA
Posts: 4,234
Default

For the A-D converter to go from 14 bits to 28 bits would require the camera to process over 16 thousand times more information. Could get a bit sluggish, but it wouldn't get any more DR from the sensor. Dynamic Range is a measure of the difference between the smallest signal recorded by the sensor, to the largest. The voltages are on the order of a volt maximum. Using the analogy of a photon bucket, the buckets can't get any deeper, so it is the noise floor of the sensor which really determines the DR.
As to the idea of being able to make exposure adjustable withing the SS/Aperture/Sensitivity parameters, many, if not all, cameras do this with SS. Reading the detailed exif info from a number of images form different cameras in auto exposure, I have seen shutter speeds vary considerably from the basic settings. An exif reader such as PhotoMe will show (with some cameras) the exact SS used, as well as the standard SS which is the one normally shown. This could be something on the order of a standard SS of 1/125s and an actual of 1/138.5s The shutters of P&S cameras are pretty much all electronic, and DSLRs are electronically controlled.

brian
VTphotog is offline   Reply With Quote
Old Feb 10, 2012, 2:56 AM   #6
Senior Member
 
MartinSykes's Avatar
 
Join Date: Feb 2010
Location: Cheshire, UK
Posts: 854
Default

Quote:
Originally Posted by VTphotog View Post
For the A-D converter to go from 14 bits to 28 bits would require the camera to process over 16 thousand times more information.
I don't follow - why isn't it just double?

My thinking on more range is that when you take a shot that has a blown highlight it's simply because the camera couldn't count any higher. I think the camera can already capture more range - it just can't record it.

Yould still need a shutter for artistic reasons - controlling the amount of blur - but ISO just seems like an unnecessary limitation left over from the days of film.

In more detail, here's how I think it would work. Say you currently take a bracketed set at +-2ev so 1/15s, 1/4s and 1s. We all agree the resultant HDR has more dynamic range. Now do the same photo with a 28-bit buffer. You just take the 1s exposure which now doesn't blow the highlights and multiply the results by whatever value is necessary to get your range of values into your final 14-bit raw or 8-bit jpeg.

I'm not sure about the sensor 'photon buckets' getting full. If I take a 1600ISO shot they might get full (blown) but the same shot at 100ISO would look fine. That suggests that the sensor can handle the number of photons coming in because it's the same each time but that something somewhere else in the processing is struggling. Isn't it just that at 1600ISO the photon count in the highlihgts is multiplied up beyond the 12- or 14-bit limit? If that's the case though, it really just needs the camera to take the basic 100ISO shot and then pick a multiplier that doesn't clip the highlights.
__________________
AutoHDR - Free HDR software for Windows at http://www.autohdr.co.uk on Facebook at http://www.facebook.com/home.php?sk=...36045126467361
My Flickr stream: http://www.flickr.com/photos/martinsykes
Panasonic DMC-FZ38, Sony ­α580

Last edited by MartinSykes; Feb 10, 2012 at 3:01 AM.
MartinSykes is offline   Reply With Quote
Old Feb 10, 2012, 5:36 AM   #7
Senior Member
 
Join Date: Apr 2010
Posts: 584
Default

Quote:
Originally Posted by MartinSykes View Post
I don't follow - why isn't it just double?
In linear digital sampling systems, one bit is twice the voltage,
four times the power or if you prefer 6dB (decibels).

Quote:
Now do the same photo with a 28-bit buffer.
You can't have 28bits worth of DR just by wishing for it. Even if
you could, there is no technology which could display the
recorded image. If there was, you would have to wear a welding
mask to safely look at a picture of someone welding.

Just to put things in perspective, the entire UK national
electricity grid is currently running at 54,000,000,000 watts.
http://www.nationalgrid.com/uk/Elect...d/demand24.htm

The small LED lamp on my bicycle takes 1W. The power
difference between these two extremes is :
10log (54000000000) = 107dB or about 18 bits.
corkpix is offline   Reply With Quote
Old Feb 10, 2012, 6:53 AM   #8
Senior Member
 
MartinSykes's Avatar
 
Join Date: Feb 2010
Location: Cheshire, UK
Posts: 854
Default

Here's a possible real-world example:

I want to take a shot of the stars with a 8s exposure at 3200ISO. There's a full-moon in the frame which will obviously be blown out and ruin the shot. At 1/200s and ISO 100 The moon might be OK but the stars would be under exposed. That's maybe 16 stops or more in a single shot? A sunny garden shot through a window from indoors has a huge range so there are definitely real-world examples that apply.

What actually happens in the camera that causes the moon to be blown? What is it in the camera that can't count high enough and why can't we just make it count higher? What is the ISO setting actually doing to the camera that requires it to be determined *before* the shot? If the camera could count much higher could ISO become something else that you apply to the RAW in PP, just like the WB?

The actual display of the image is a separate issue which we already have now with most cameras that can already capture more than we can display.
__________________
AutoHDR - Free HDR software for Windows at http://www.autohdr.co.uk on Facebook at http://www.facebook.com/home.php?sk=...36045126467361
My Flickr stream: http://www.flickr.com/photos/martinsykes
Panasonic DMC-FZ38, Sony ­α580
MartinSykes is offline   Reply With Quote
Old Feb 10, 2012, 7:49 AM   #9
Senior Member
 
Join Date: Apr 2010
Posts: 584
Default

Quote:
Originally Posted by MartinSykes View Post

What actually happens in the camera that causes the moon to be blown?
There is a limit to the number of electrons that you can store in
the 'well'. Adding 14 bits increases that number by more than
16,000x. Even at the current state-of-the-art in sensor technology,
there are problems caused by leakage of electric charge to adjacent
pixels which leads to 'blooming', reduced resolution and inaccurate colour.
Imagine how much worse this would be if the sampled charge was
16,000 times higher.

Quote:
What is it in the camera that can't count high enough and why can't we just make it count higher?
The counting is done by an ADC. There is a limit to the DR that can
be sampled. There is no point in trying to sample at a bit depth that
is significantly greater than the sensor DR. This extends from the
noise floor at the dark end to blown highlights at the other.

Noise comes from several sources. The predictable ones are thermal
(Johnson) noise and shot noise which is proportional to electric current
in the camera's circuits.

Quote:
What is the ISO setting actually doing to the camera that requires it to be determined *before* the shot?
As I understand it, coarse adjustment of ISO setting is achieved by
changing the gain of the amplifiers between the sensor photosites
and the ADC. Fine adjustment of ISO such as the 1/3 stop increments
between the traditional 1-stop values are sometimes achieved by
manipulating the digital data after the shot is taken and sampled.

Quote:
If the camera could count much higher could ISO become something else that you apply to the RAW in PP, just like the WB?
If the camera could behave linearly up to a much higher light level,
then you could indeed record a greater dynamic range.

Noise at the low (dark) end can be reduced by reducing electric
current in the sensor. This conflicts with the market requirement
for greater speed and more pixels (CMOS circuits need more
power at higher clock speeds). The other way to reduce noise
is to cool the sensor and associated circuits to a temperature
approaching absolute zero. This isn't practical in a portable camera,
but such techniques are used in radio astronomy.

Each stage in the chain: photosite (individual CMOS or CCD pixel),
amplifier and ADC has a limited dynamic range. High quality amplifiers
and ADCs have a dynamic range of greater than 100dB (about 17 bits).
Hi-Fi audio ADCs and DACs have a bit depth of 24 bits, but in practice,
they rarely if ever achieve a DR much beyond 100dB. In practice,
several bits are effectively lost to peak headroom at one end and
noise at the other.
corkpix is offline   Reply With Quote
Old Feb 10, 2012, 8:49 AM   #10
Senior Member
 
MartinSykes's Avatar
 
Join Date: Feb 2010
Location: Cheshire, UK
Posts: 854
Default

Thanks corkypix. So could you achieve the same result by using say 1000 1/200s exposures instead of one 5s exposure and then combining them in-camera? The combined file would require a greater bit-depth but the sensor wouldn't. You might need to implement a software shutter instead of using the mechanical one but it should work. I think it should be possible to demonstrate this using a conventional DSLR - it just requires the combination of a number of identical exposures instead of bracketed exposures. I'll give it a try later today.
__________________
AutoHDR - Free HDR software for Windows at http://www.autohdr.co.uk on Facebook at http://www.facebook.com/home.php?sk=...36045126467361
My Flickr stream: http://www.flickr.com/photos/martinsykes
Panasonic DMC-FZ38, Sony ­α580
MartinSykes is offline   Reply With Quote
 
Reply


Thread Tools Search this Thread
Search this Thread:

Advanced Search

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are Off
Pingbacks are Off
Refbacks are Off



All times are GMT -5. The time now is 11:16 PM.