Nikon D80 - auto ISO

[terpeh]

1/15 is quite a slow shutter and prone to handshake. I am glad i have a f2.8 constant lense which do help a bit to increase the shutter during short zoom range, ie up to 50mm. As for long range, i rely heavily on the VR to support.

Wonder whether Nikon will come out with new firmware to fix this problem... something it is a bit irritating, very unpredictable.

 

[gooseberry]

but issit true that 1/3 stops and smaller steps of ISO is actually just software exposure corrected like what calebk has mentioned?

Actually, any ISO above the sensor's base ISO involves the use of signal amplification (be it electronic or analogue). If your sensor has a base ISO of 100, and you set ISO 400 on your camera, your sensor is not now 2 stops more sensitive (unlike in film, where an ISO 400 film is actually 2 stops more sensitive than an ISO 100 film). When you set your camera to ISO 400, and you expose for ISO 400, you are actually underexposing your sensor by 2 stops, and the camera does a 2 stop signal amplification to get the "correct" exposure.

 

[Scaglietti]

1/15 is quite a slow shutter and prone to handshake. I am glad i have a f2.8 constant lense which do help a bit to increase the shutter during short zoom range, ie up to 50mm. As for long range, i rely heavily on the VR to support.

Wonder whether Nikon will come out with new firmware to fix this problem... something it is a bit irritating, very unpredictable.

Firmware to fix which problem?

BC

 

[posez]

Firmware to fix which problem?

BC

I think he's talking about the "overexposure" problem. That's a discussion for another thread, mya fault for mentioning it

 

[Zaknafein]

but issit true that 1/3 stops and smaller steps of ISO is actually just software exposure corrected like what calebk has mentioned?

Actually, any ISO above the sensor's base ISO involves the use of signal amplification (be it electronic or analogue). If your sensor has a base ISO of 100, and you set ISO 400 on your camera, your sensor is not now 2 stops more sensitive (unlike in film, where an ISO 400 film is actually 2 stops more sensitive than an ISO 100 film). When you set your camera to ISO 400, and you expose for ISO 400, you are actually underexposing your sensor by 2 stops, and the camera does a 2 stop signal amplification to get the "correct" exposure.

actually what i meant was not the sensor's signal amplification, but rather, does D80 uses "software manipulation" to simulate signal amplification for ISO like 125, 500, 640 etc....

 

[gooseberry]

actually what i meant was not the sensor's signal amplification, but rather, does D80 uses "software manipulation" to simulate signal amplification for ISO like 125, 500, 640 etc....

Yes, and that is what I am saying... any ISO above the base ISO involves manipulation to amplify the signal. Any ISO above the base ISO is not a "true" ISO (whether it be 1/3, 1/2 or full stops), it is all signal amplification.

You can do a little experiment to check for yourself on your D80. Set your camera to RAW and ISO to 400 on a tripod. Meter a scene and set your aperture and shutter speed manually to "correctly" expose for ISO 400. Take a shot. Now, leaving your aperture and shutter speed the same, change your ISO to 100. Take another shot. This second shot will be underexposed by 2 stops compared to your first shot. Now using a RAW converter that allows you to adjust exposure EV (I'd recommend Nikon Capture at least, there are others you can use too, but don't use Adobe Camera Raw (ACR) as it's exposure adjustments seems to introduce more noise - besides ACR is not very optimised for Nikon cameras), adjust the second shot up by 2 stops and then tell me whether the two photos look the same.

 

[averynkh]

when i read through this thread only i kow there is auto selection for iso, then when i get home, i tried to set to auto, but i couldnt do it at all.. auto selection is off.. i read the guide book, it doesnt really figure oout wat to do.. only auto mode can set iso to auto but without max iso sensitivity and min shutter option too... guys... pls help me how to do auto iso in M, A or S mode.. coz i hate my pic taken in event...

 

[averynkh]

i think i got it.. shiting finally... let me try

 

[wuyongzheng]

Yes, and that is what I am saying... any ISO above the base ISO involves manipulation to amplify the signal. Any ISO above the base ISO is not a "true" ISO (whether it be 1/3, 1/2 or full stops), it is all signal amplification.

You can do a little experiment to check for yourself on your D80. Set your camera to RAW and ISO to 400 on a tripod. Meter a scene and set your aperture and shutter speed manually to "correctly" expose for ISO 400. Take a shot. Now, leaving your aperture and shutter speed the same, change your ISO to 100. Take another shot. This second shot will be underexposed by 2 stops compared to your first shot. Now using a RAW converter that allows you to adjust exposure EV (I'd recommend Nikon Capture at least, there are others you can use too, but don't use Adobe Camera Raw (ACR) as it's exposure adjustments seems to introduce more noise - besides ACR is not very optimised for Nikon cameras), adjust the second shot up by 2 stops and then tell me whether the two photos look the same.

I disagree. If it is a simple software amplification, the noise would be amplified as well as the signal. i.e. the signal noise ratio would be the same as long as there is no overexposure. In fact it's not.

I did a noise test on my d80 recently. I have put the result at http://www.comp.nus.edu.sg/~wuyongzh/my_doc/noise/

The two cases DSC_6987.NEF and DSC_6989.NEF are same experiment as what you have described right? Their signal noise ratio are different! ISO400's SNR is larger, so it's better to shoot at ISO400 instead of ISO100 and adjust the exposure.

 

[averynkh]

hey guys, y doesnt this happen on me? :dunno:

i set max sensitivity to 800, min shutter to 1/30
then iso at 400.. A mode... without flash at first
i test in my room, fluorescent lamp on the ceiling
anyway, no matter with or without flash
the camera still shake, so the pic a little blur :cry:
if with flash, can i use rear flash or normal mode?

 

[Override2Zion]

I disagree. If it is a simple software amplification, the noise would be amplified as well as the signal. i.e. the signal noise ratio would be the same as long as there is no overexposure. In fact it's not.

I did a noise test on my d80 recently. I have put the result at http://www.comp.nus.edu.sg/~wuyongzh/my_doc/noise/

The two cases DSC_6987.NEF and DSC_6989.NEF are same experiment as what you have described right? Their signal noise ratio are different! ISO400's SNR is larger, so it's better to shoot at ISO400 instead of ISO100 and adjust the exposure.

I beg to differ, the test might not be absolutely accurate, what you've done is a comparison of SNRs between images extracted at the end stage. Raw data from the sensors are interpolated and goes through a whole lot of digital manipulation by the camera's image processing engine. We can't be sure if turning off NR in the D80 would result in bypassing the NR algorithm totally.

 

[Override2Zion]

hey guys, y doesnt this happen on me? :dunno:

i set max sensitivity to 800, min shutter to 1/30
then iso at 400.. A mode... without flash at first
i test in my room, fluorescent lamp on the ceiling
anyway, no matter with or without flash
the camera still shake, so the pic a little blur :cry:
if with flash, can i use rear flash or normal mode?

Under normal circumstances you'd be using the flash in normal mode. If you wanna get creative with slow shutters and motion blurs, you'd be looking into the rear/front curtain flash modes. Btw, fluorescent tubes aren't very good for testing since they are flickering consistently as the AC cycles. Try setting your minimum shutter speed to 1/60 instead, should help further reduce handshake blur.

 

[wuyongzheng]

Raw data from the sensors are interpolated and goes through a whole lot of digital manipulation by the camera's image processing engine. We can't be sure if turning off NR in the D80 would result in bypassing the NR algorithm totally.

Do you mean color interpolation (demosaicing)? That's for jpeg results, not raw. I used the green channel in raw for the test.

NR doesn't matter about raw. (Long exposure NR (aka. dark frame subtraction) does, but that's a different story.) Not only NR, color saturation, D-lighting and color space don't matter either. I have explained this already in the test result page.

I think the only things matter about raw are analog-digital converter and lossy NEF compression. D80's AD converter only gives 12 bits. I can do nothing about it. I have some discussion on lossy NEF compression on the result page.

Is there any "digital manipulation by the camera's image processing engine" I messed here?

 

[Override2Zion]

Is there any "digital manipulation by the camera's image processing engine" I messed here?

My point is no outsider know's how "raw" is RAW. Please take a look at some test done by others.

http://forums.dpreview.com/forums/read.asp?forum=1034&message=29131601

http://www.nikonians.org/forums/dcboard.php?az=show_topic&forum=147&topic_id=53340

There are endless arguments about how raw is really raw. Does "straight out of the sensor" still stand? Perhaps only Nikon will know. Ain't a big deal to me though, so long as I can still keep on making pictures that I like.

Btw, sidetrack a little. Why did you choose to make the reference images from reflected light? Shoot at the wall/floor, are those surfaces totally flat, how to ensure if the light intensity is consistent? Why not perform the test again with your lens cap on and seal your viewfinder, you'll eliminate most of the above-mentioned factors that may introduce inconsistency. No problems of signal saturation either. You've got some interesting research going on there, Happy Engineering!

 

[wuyongzheng]

My point is no outsider know's how "raw" is RAW. Please take a look at some test done by others.

Agree. There maybe some unknown processing before raw. That's why we do testing.

http://forums.dpreview.com/forums/read.asp?forum=1034&message=29131601

D90's shutter is 1/400 but D300's shutter is 1/500. I think that might be the reason D300 is noisier. Think of the extreme case when the shutter is 1/500000, the result would be dominated by noise already (signal almost goes to 0).

Btw, sidetrack a little. Why did you choose to make the reference images from reflected light? Shoot at the wall/floor, are those surfaces totally flat, how to ensure if the light intensity is consistent? Why not perform the test again with your lens cap on and seal your viewfinder, you'll eliminate most of the above-mentioned factors that may introduce inconsistency. No problems of signal saturation either. You've got some interesting research going on there, Happy Engineering!

Note I'm shooting the wall without any lens. That should make sure each pixel gets the same amount of light. Actually there is another problem, shade. The corners may get darker. But since I only use the center 256x256 region, the shade effect shouldn't be strong.

Noise comes from many sources. Testing the noise at zero light covers one of the sources which is probably caused by heat. It's especially important to astronomy photographer. But I'm not very interested in it as I seldom do long exposure. I'm more interested in high ISO noise in normal lighting condition and normal shutter speed.

 

[Override2Zion]

Noise comes from many sources. Testing the noise at zero light covers one of the sources which is probably caused by heat. It's especially important to astronomy photographer. But I'm not very interested in it as I seldom do long exposure. I'm more interested in high ISO noise in normal lighting condition and normal shutter speed.

I dun quite get what you mean by shooting without lenses will expose them to even amounts of light, ambient lighting condition changes as well. If ambient light is even and consistent, shade is not a problem either since the light falloff should also be consistent right? You dun have to do long exposures to verify if the ISO noise, in fact you won't need the image to be exposed properly. You could still carry out the test with manual control of shutter speed, ISO and aperture. To perform a fair test, we would want to eliminate as many variables as we can.
Understand where you're coming from though. Keep up the good work though, many will benefit from your findings

 

[wuyongzheng]

I dun quite get what you mean by shooting without lenses will expose them to even amounts of light, ambient lighting condition changes as well. If ambient light is even and consistent, shade is not a problem either since the light falloff should also be consistent right? You dun have to do long exposures to verify if the ISO noise, in fact you won't need the image to be exposed properly. You could still carry out the test with manual control of shutter speed, ISO and aperture. To perform a fair test, we would want to eliminate as many variables as we can.

Even though the brightness of the wall is not even, the light reflected by the wall will evenly distribute on the sensor. If you don't understand why, just think of shooting a laser pen on the wall. The brightness of the wall is extremely uneven, right? But the light reflected by the wall can evenly reach a piece of paper in front of the wall. I.e. you don't see red dot on the paper, right?

Regarding to your suggestion, I think (haven't tested out) if you keep the lens/body cap on and take a picture with normal shutter speed (< 1/10s), the picture will come out totally black no matter how high the ISO is. That's why all the zero-light tests (including the 2nd link you gave) use long exposure. In addition, this kind of noise can be effectively reduced by dark frame subtraction (or long exposure NR called by nikon). In the 2nd link you gave, long exposure NR is cause of for the sudden noise drop. Because long exposure NR only takes place if shutter exceeds certain threshold. The threshold is the sudden drop point.

 

[Override2Zion]

But the light reflected by the wall can evenly reach a piece of paper in front of the wall. I.e. you don't see red dot on the paper, right?

You mentioned that the shaded (light falloff) area be excluded and only samples from the center portion of the sensor is used. Suggestion of shooting with lens cap on eliminates this ambiguity (shade) and also increases the area of the samples that are made available to you. In your case, if ambient light intensity remains extremely constant for all your shots, you'd only be measuring poisson noise in addition to sensor read noise.

if you keep the lens/body cap on and take a picture with normal shutter speed (< 1/10s), the picture will come out totally black no matter how high the ISO is.

This statement is not true, there is still noise embedded within. What you've measured with lenses cap off is poisson noise on top of read noise. Your attempt of the test with constant light does not reflect the worst case scenario, since poisson noise gets more apparent as available light reduces. I understand your point about shooting under normal lighting/shutter conditions though but what defines "normal lighting condition"? When shooting outdoors and indoors, exposure parameters can vary quite a lot. Such situations are still considered "under normal light/shutter". Read the part about "sensor read noise" from the following article as well. May serve as good reference in your research, rather informative.
http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/index.html

In addition, this kind of noise can be effectively reduced by dark frame subtraction (or long exposure NR called by nikon).

Your test is suppose to be based on Nikon NEF Raws right? NR be it single or long exposures shouldn't be active like you mentioned, data is straight out of the sensors for RAW. NR algorithms like dark frame/pattern noise subtraction, CDS, flat field, etc should not have been applied here by the camera. Unless you're only stating facts for reference purposes.

After all, any form of such measurements are not really important to most photographers. Might be interesting for the sake of engineering though. In most real life photography applications, timing is critical, sometimes things happen so fast that we dun really have much time to think about what settings to use to minimize noise, as a matter of fact, some photographers wants noise in their pictures.

 

[averynkh]

i went to elva autograph session today, i tried to use A mode with auto iso which max sensitivity set at 800 and max shutter at 1/80, then original iso at 100, but i dun find any good points as a lot pp said this is very smart function in d80... i still get underexposore shots...

here you go...
http://www.sadhuism.com/pic/autoiso1.JPG
http://www.sadhuism.com/pic/autoiso2.JPG

 

[gooseberry]

I disagree. If it is a simple software amplification, the noise would be amplified as well as the signal. i.e. the signal noise ratio would be the same as long as there is no overexposure. In fact it's not.

I did a noise test on my d80 recently. I have put the result at http://www.comp.nus.edu.sg/~wuyongzh/my_doc/noise/

The two cases DSC_6987.NEF and DSC_6989.NEF are same experiment as what you have described right? Their signal noise ratio are different! ISO400's SNR is larger, so it's better to shoot at ISO400 instead of ISO100 and adjust the exposure.

Been away travelling so missed this.

It's not just software amplification (what some refer to as digital shift) but also analogue signal amplification via amplifiers.

Haven't looked at your results and method in detail, but if I am reading your table correctly, the SNR for DSC_6987.NEF is 17dB and for DSC_6989.NEF it is also 17dB - which is the same is it not ? (actually 16.6dB for one and 16.5dB for the other)