Garbz
No longer a newbie, moving up!
- Joined
- Oct 26, 2003
- Messages
- 9,713
- Reaction score
- 203
- Location
- Brisbane, Australia
- Website
- www.auer.garbz.com
- Can others edit my Photos
- Photos NOT OK to edit
This is something that pops up constantly. The idea that if you shoot to the right and then adjust your image in post processing it acts as a sort of a noise reduction.
This irked me as it goes against everything I could think of with using a camera. Mainly what got me is the idea of overexposing. If you can over expose then you can also just simply drop your ISO. I have long believed That the step change in noise at the high end of the ISO scale of cameras is very severe indeed. The noise increase is somewhat linear towards the bottom scale from the common graphs, but seems to rise exponentially after a certain point, and that is usually the point where noise starts to matter in an image.
After a while of trying to justify to myself that reducing the brightness of an image constitutes less of an improvement than reducing the ISO I gave up, and grabbed my camera. I present to you the results of my test:
Below are two images of my slightly cream wall, on the edge is a black picture frame, and everything is nicely out of focus to make damn certain that noise isn't mistook for detail.
The first image was shot at ISO800 1/6th f/5.6. The second image was shot at ISO1600 1/6th f/5.6 and reduced in brightness in Lightroom 1.0EV. A quick sanity check was done to check the brightness of the image so it's unchanged and over a 31x31 pixel average which gave:
ISO800 rgb(147,148,130)
ISO1600 rgb(146,147,131)
Wait ok firstly the colours no longer match! I want to go back and give this a go with a colour checker card to see how bad it is, but really right now I will assume that this slight difference isn't field relevant.
The images: ISO800:
ISO1600 -1EV adjusted:
Results:
Ok so the ISO1600 shot did get better when reduced. That was expected, but the point is to find out how much it got better by. To do this the left quarter of the frame was selected in photoshop. The histogram was set to extended information on the colour channel and the following standard deviations were noted:
ISO800 vs ISO1600
Black area:
4.52 vs 4.79
Light area:
8.31 vs 8.40
Then the measurements were repeated with the histogram set to luminosity:
ISO800 vs ISO1600
Black area:
3.38 vs 3.64
Light area:
2.48 vs 2.89
Now there is an uncertainty in that I'm not sure how very "scientific" the results from photoshop could be trusted. I'm inclined to believe they represent the images, but I'm not sure how much meaning the numbers have.
However I satisfied my own belief that if you're in a situation where it is possible to shoot to the right and then underexpose in post, you may well be better served dropping your ISO to just simply expose correctly. Noise wise the differences favour the lower ISO shot, but looking at the images it really seems like this discussion isn't worth having
Now one other thing I noted earlier is that the noise response of cameras typically change depending how close to the end of the ISO scale you actually get. This result here for the Nikon D200 between ISO800-1600 will likely not match the results on a Nikon D700 between ISO800-1600 (and the overexpose underprocess trick may work), however I would bet the results would match a D700 ISO12800-25600.
Let's hear from other people. What are your experiences? Perhaps if someone else could repeat this test on a different camera?
This irked me as it goes against everything I could think of with using a camera. Mainly what got me is the idea of overexposing. If you can over expose then you can also just simply drop your ISO. I have long believed That the step change in noise at the high end of the ISO scale of cameras is very severe indeed. The noise increase is somewhat linear towards the bottom scale from the common graphs, but seems to rise exponentially after a certain point, and that is usually the point where noise starts to matter in an image.
After a while of trying to justify to myself that reducing the brightness of an image constitutes less of an improvement than reducing the ISO I gave up, and grabbed my camera. I present to you the results of my test:
Below are two images of my slightly cream wall, on the edge is a black picture frame, and everything is nicely out of focus to make damn certain that noise isn't mistook for detail.
The first image was shot at ISO800 1/6th f/5.6. The second image was shot at ISO1600 1/6th f/5.6 and reduced in brightness in Lightroom 1.0EV. A quick sanity check was done to check the brightness of the image so it's unchanged and over a 31x31 pixel average which gave:
ISO800 rgb(147,148,130)
ISO1600 rgb(146,147,131)
Wait ok firstly the colours no longer match! I want to go back and give this a go with a colour checker card to see how bad it is, but really right now I will assume that this slight difference isn't field relevant.
The images: ISO800:
ISO1600 -1EV adjusted:
Results:
Ok so the ISO1600 shot did get better when reduced. That was expected, but the point is to find out how much it got better by. To do this the left quarter of the frame was selected in photoshop. The histogram was set to extended information on the colour channel and the following standard deviations were noted:
ISO800 vs ISO1600
Black area:
4.52 vs 4.79
Light area:
8.31 vs 8.40
Then the measurements were repeated with the histogram set to luminosity:
ISO800 vs ISO1600
Black area:
3.38 vs 3.64
Light area:
2.48 vs 2.89
Now there is an uncertainty in that I'm not sure how very "scientific" the results from photoshop could be trusted. I'm inclined to believe they represent the images, but I'm not sure how much meaning the numbers have.
However I satisfied my own belief that if you're in a situation where it is possible to shoot to the right and then underexpose in post, you may well be better served dropping your ISO to just simply expose correctly. Noise wise the differences favour the lower ISO shot, but looking at the images it really seems like this discussion isn't worth having
Now one other thing I noted earlier is that the noise response of cameras typically change depending how close to the end of the ISO scale you actually get. This result here for the Nikon D200 between ISO800-1600 will likely not match the results on a Nikon D700 between ISO800-1600 (and the overexpose underprocess trick may work), however I would bet the results would match a D700 ISO12800-25600.
Let's hear from other people. What are your experiences? Perhaps if someone else could repeat this test on a different camera?