GH2 Read/Dark Noise plots aplenty

Started Dec 20, 2011 | Discussions thread
ForumParentFirstPreviousNextNext unread
Flat view
kenw
Veteran MemberPosts: 4,194Gear list
Like?
GH2 Read/Dark Noise plots aplenty
Dec 20, 2011

So there are probably all of two people on the entire forum that care at all about this, so feel free to close this thread and go take some photos instead. And, by the way, this is Detail Man's fault - I should be working or photographing instead of making boring plots.

The take-away from this, if there is one at all, is that ISOs 250, 500, 1000, 2000... have excessive read (shadow) noise and are probably best avoided. The "best" ISOs to use are 160, 320, 640, 1250... and so on. Also, for long exposures the "Long Exposure NR" function does not work properly at ISO1000.

Also, not really important but perhaps of academic interest is that ISO800 is the last "real" ISO, the higher ISOs are digital gains. This is not unusual for most cameras, there is no advantage to analog gain once you are just a few stops above base ISO.

And now on to all the boring plots to back that up...

First up, RMS noise in dark exposures (essentially read noise) at all ISOs for 1/4000 and 2 sec exposures with various NR settings. These are RAW counts - no RAW converter was used, read into Matlab from a DNG file.

From it we can see that with NR+2 set we will get RAW NR starting at ISO 800. Also we see the set of ISOs 250, 500, 1000 and so on have worse noise than an ISO 1/3 EV higher (320, 640, 1250). Not surprisingly we see that at longer exposures we get more noise, probably from dark current. Lastly we see that Long Exposure NR actually does help (this is in camera dark frame subtraction) but at ISO1000 it is not working properly and at low ISOs and very high ISOs the benefit is small.

Please note this is plotted against "Stated ISO" and it is possible that the actual RAW ISO is slightly different than stated at the 1/3 EV steps which could nudge the points a little to the left or right. So don't obsess too much. That said, we can see those 250, 500, 1000... ISOs are actually worse than the ISO series 1/3 EV higher.

For the record, what is plotted above is:

mean(mean((cfa - 16).^2)))^0.5

Where cfa is the entire 16MP RAW array (it is 2D, that's why mean is called twice). The lowest count is always 16 and is used as the black level, that's why I'm subtracting 16.

Next are four channel histograms, separate curves for R, B, G1 and G2 in each plot. I have no idea which is which, as you can see the four channels have nearly identical histograms. A histogram value of 10e-2 represents a zero count, I did that so zero count bins would still plot and be obvious. Note that the black level is always 16 regardless of ISO or other settings. Also note that the distribution is clipped at the black level, that is the camera does not record the bottom half of the distribution. You can see this as the black level bin is significantly more full than the higher bins. A spot check shows in general the black level bin contains the same count as all the other bins combined. This indicates the camera is setting the black level at the mean of the distribution and then clipping the bottom half of the distribution.

In this pair we can see digital gain being applied at ISO1000. Notice all the zero count bins in the ISO1000 plot where as the ISO800 histogram is continuous. All ISOs 1000 and above show digital gain and zero count bins, all ISOs 800 and below show no digital gain and have no zero count bins.

Here we see the same pair but with NR+2. Notice the dramatic change in the shape of the histogram at the low ADC values. Also note that the empty bins at ISO1000 are gone. This is because the NR is being applied after the digital gain is applied so those bins are now filling in from NR averaging.

Finally, here is Long Exposure NR (in camera dark frame subtraction) at work for ISOs 640, 800 and 1000. Note that it is quite effective at 640 and 800 but doesn't improve much at all for 1000. I suspect there is an error in the implementation relating to when digital gain is applied and the dark frame subtraction performed at this ISO - but I can only guess.

Note also that this Long Exposure NR from dark frame subtraction is clearly very different from the NR+2 RAW NR - look closely at the left most bins and note that the shape of the distribution isn't altered the way the NR+2 RAW NR is. Instead the subtraction is removing the bump further to the right in the distribution as well as the outliers far to the right. This is exactly what dark frame subtraction is suppose to do.

Clear as mud? No worries, ignore the whole post and take some pictures instead.
--
Ken W
See plan in profile for equipment list

 kenw's gear list:kenw's gear list
Sony RX100 Panasonic Lumix DMC-G1 Olympus OM-D E-M5 Panasonic Lumix DMC-GM1 Panasonic Lumix G Vario 14-45mm F3.5-5.6 ASPH OIS +25 more
Panasonic Lumix DMC-G2 Panasonic Lumix DMC-GH2
If you believe there are incorrect tags, please send us this post using our feedback form.
ForumParentFirstPreviousNextNext unread
Flat view
Post (hide subjects)Posted by
ForumParentFirstPreviousNextNext unread
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark post MMy threads
Color scheme? Blue / Yellow