Ad
Ad
Ad
Pages: [1]   Bottom of Page
Print
Author Topic: Push ISO or not? Canon vs Nikon vs Sony  (Read 6593 times)
Guillermo Luijk
Sr. Member
****
Offline Offline

Posts: 1299



WWW
« on: March 13, 2011, 02:09:31 PM »
ReplyReply

Getting a proper RAW exposure at base ISO is the ideal situation to improve image quality (i.e. reduce noise).

However, in low light conditions, even using the slowest shutter speed and the widest aperture we can afford in our application, we may not get a proper RAW exposure at the lowest ISO setting. In this situation, pushing ISO is recommended to get a better RAW exposure and hence lower noise, but the convenience to do it is not equally critical in all brands.

I have compared the ISO behaviour of the Canon 5D Mark II, the Nikon D700 and the Pentax K5 (with a Sony sensor). Most Canon sensors behave similarly, so as most Nikon sensors, so this can also be considered a brand comparision. The K5 sensor is the same Sony sensor found in the Nikon D7000 and the Sony A380.

Using noise data from Sensorgen.info for the three cameras under test, collected from DxOMark SNR plots and normalised to an output resolution of 12,7Mpx to make it a fair comparision, the following plots are obtained (the black line represents the improvement in SNR at each ISO increment. The line is plotted gray once improvement becomes negligible):


The improvement in SNR after a 5 stops ISO increment is very different for the 3 types of sensors:
  • Canon: SNR improves up to 2.1EV, up to ISO1600, and specially at the lowest ISO values
  • Nikon: SNR improves up to 1EV and up to ISO800
  • Sony: SNR improves up to 1/3EV and progressively at all ISO values

The conclusion for those 3 different behaviours would be to use each camera in a different way according to:
  • Canon: it's strongly recommended to push ISO on a Canon camera as soon as RAW underexposure may happen at base ISO, because improvement in noise will be very important. Pushing ISO beyond ISO1600 becomes useless.
  • Nikon: it's not so critical to push ISO to avoid RAW underexposure as in Canon, but still recommended since the there will still be a clear noise reduction. Pushing ISO beyond ISO800 becomes useless.
  • Sony: noise reduction through ISO is so slight, that it is not worth to push ISO even if RAW underexposure may take place. By staying at base ISO in all cases, we'll save highlights information (in case the scene has got them) and will not have noisier shadows.

The Sony sensor is close to an ISOless sensor (the first time I heard this term was from Emil Martinec). This means it could be adjusted at base ISO in any situation, and just concentrate in choosing the most adequate shutter and aperture. Unluckily this will make the camera display look very dark in some situations; camera manufacturers could allow to adjust the display brightness or just consider the ISO setting a metadata parameter that only affects the camera display, not altering RAW data.
_______

I also plotted the dynamic range of the 3 tested cameras (12dB SNR criteria, 12,7Mpx output resolution) vs the real ISO of each sensor. The APS-C sized Pentax clearly beats the two FF cameras at low ISOs, behaves similarly at ISO400 and looses at high ISOs. No clear winner among the FF cameras; the Nikon D700 is slightly better than the 5D2 at low ISOs (Nikon's ISO200 even beats Canon's ISO100), but we must also consider Canon's higher pixel count.


Regards
« Last Edit: March 14, 2011, 01:10:38 PM by Guillermo Luijk » Logged

bjanes
Sr. Member
****
Offline Offline

Posts: 2835



« Reply #1 on: March 14, 2011, 10:49:27 AM »
ReplyReply

Guillermo,

An excellent analysis, but the sensors of the D700 and the closely related D3 have increased read noise at low ISO and the DR at low ISO is limited by increased read noise as indicated by flattening of the DR curve towards the left. The DR plot of the D7000 is nearly linear and it too is close to being Emil's ISO less camera. I understand that the same considerations apply to the Hassselblad MFDBs.

Regards,

Bill
Logged
MrSmith
Sr. Member
****
Offline Offline

Posts: 916



WWW
« Reply #2 on: March 14, 2011, 02:26:11 PM »
ReplyReply

the 5DII's native iso is 160, and i only shoot at multiples of that for best results (320, 640 etc).
i find that's the best way of reducing noise, especially the pattern noise in the blacks, would your test results be different if the ideal iso's are used?
Logged
ejmartin
Sr. Member
****
Offline Offline

Posts: 575


« Reply #3 on: March 14, 2011, 02:52:23 PM »
ReplyReply

the 5DII's native iso is 160, and i only shoot at multiples of that for best results (320, 640 etc).
i find that's the best way of reducing noise, especially the pattern noise in the blacks, would your test results be different if the ideal iso's are used?

No, because these ISO's are no different than the 1/3 stop higher ISO's they are derived from:

http://www.luminous-landscape.com/forum/index.php?topic=51782.msg427769;topicseen#msg427769

The 5D2's "native" ISO is not 160; I'm not sure why you think it is.  The ISO values that correspond to analogue signal amplification are 100-200-400-800-1600-3200.  All other ISO's are derived from these analogue ones by digital manipulation after signal quantization.
« Last Edit: March 14, 2011, 02:55:32 PM by ejmartin » Logged

emil
MrSmith
Sr. Member
****
Offline Offline

Posts: 916



WWW
« Reply #4 on: March 15, 2011, 03:52:35 PM »
ReplyReply

i'm not a techie but i was looking at this chart:


after running into pattern noise issues with the very first job i shot of a white road bike wheel against black, i ended up masking off the black area's and overlaying with a generated background to get rid of the pattern noise (in fairness it probably wouldn't reproduce in print but didn't want the client to see it on their screen)
i did a few simple tests and found the 160 to give better files than 100 (as did the multiples of 160 than the inbetween iso's)
most of my shooting is done tethered on a tripod with critical exposures just shy of clipping.

i'm happy to be proved wrong but i can only go on what i see on screen.
Logged
ejmartin
Sr. Member
****
Offline Offline

Posts: 575


« Reply #5 on: March 15, 2011, 04:19:57 PM »
ReplyReply

i'm not a techie but i was looking at this chart:


after running into pattern noise issues with the very first job i shot of a white road bike wheel against black, i ended up masking off the black area's and overlaying with a generated background to get rid of the pattern noise (in fairness it probably wouldn't reproduce in print but didn't want the client to see it on their screen)
i did a few simple tests and found the 160 to give better files than 100 (as did the multiples of 160 than the inbetween iso's)
most of my shooting is done tethered on a tripod with critical exposures just shy of clipping.

i'm happy to be proved wrong but i can only go on what i see on screen.

The 5D2 obtains ISO 160 by taking an image at ISO 200 and multiplying all the digitized values by 0.8.  You could do the same by shooting at ISO 200 and dialing in -1/3 EV of EC in the raw converter.  Noise in raw levels is lower because of the 0.8 multiplier, not because of any intrinsic advantage of ISO 160 since it is the same as ISO 200 as far as the camera electronics is concerned.  ISO 160 on the 5D2 is ISO 200 with some trivial EC post-capture applied in the camera instead of the converter.  It can't have any more DR than 200 because it is 200.

You could be right that ISO 100 has more pattern noise relative to exposure than ISO 160 (=200), if you have the headroom not to clip desired highlights at 200.
« Last Edit: March 15, 2011, 04:22:21 PM by ejmartin » Logged

emil
dosdan
Newbie
*
Offline Offline

Posts: 4


« Reply #6 on: March 19, 2011, 03:40:14 AM »
ReplyReply

The K5 sensor is the same Sony sensor found in the Nikon D7000 and the Sony A380.

Hello, Guillermo. I've already been looking at the implications of your work. Since the read nose in the Sony IMX071 sensor used in the K-5, D7000, A580 & SLT A55 is so low, it gets closer than any other DSLR sensor yet to permitting "ISOless" operation (actually, operation at the fixed base ISO, relying not on variable analogue amplification before the ADC, but on digital multiplication/amplification in PP). What would ISOless/Auto EV operation mean? See here for a discussion:

The ISOless Camera

Dan.
« Last Edit: March 19, 2011, 05:31:03 AM by dosdan » Logged
Pages: [1]   Top of Page
Print
Jump to:  

Ad
Ad
Ad