fredmiranda.com Login
This topic is in the 2010 archive. You can read it and like its posts here. It is read only.

Archive 2010Noise, Interpolation, and Comparison 101

  
 
David Manning
Offline

p.1 #1Noise, Interpolation, and Comparison 101
I compared my 1D Mark IV to my original 5D at ISO 3200. This is a 100% comparison from Lightroom 3 beta with no noise reduction. Which camera is on which side?

http://www.davidmanningphotography.com/Mark-IV/MarkIVNoise.jpg

David

Edited on Feb 02, 2010 at 08:51 PM · View previous versions
Feb 02, 2010 at 08:39 PM
M Vers
Offline
Details
p.1 #2Noise, Interpolation, and Comparison 101
Since the MKIV exhibits little to no chroma noise and or blotching at this ISO I'd say the MKIV is on the right. But if you want to judge whether or not its the "king" I think you'll have to test it against a worthy opponent (5DII, D3s etc etc)

Edited on Feb 02, 2010 at 09:09 PM · View previous versions
Feb 02, 2010 at 08:41 PM
David Manning
Offline
Details
p.1 #3Noise, Interpolation, and Comparison 101
If you said right, you'd be wrong!! The Mark IV is on the left! Imagine my surprise. Don't get me wrong, I love this camera.

Edited on Feb 02, 2010 at 08:52 PM · View previous versions
Feb 02, 2010 at 08:47 PM
sivrajbm
Offline
Details
p.1 #4Noise, Interpolation, and Comparison 101
Canon ISO yes, Nikon D3s is the KING of camera ISO...
As my grandson says Sorry bout that...
Feb 02, 2010 at 08:51 PM
M Vers
Offline
Details
p.1 #5Noise, Interpolation, and Comparison 101
32067dlm wrote:
If you said right, you'd be wrong!!


Hmm, what do the results look like if run through DPP? Also, what were the in camera settings? The result seems uncharacteristic for both the MKIV, at least from what I have seen so far, and the 5D.
Feb 02, 2010 at 08:51 PM
brainiac
Offline
[X]
Details
p.1 #6Noise, Interpolation, and Comparison 101
32067dlm wrote:
This is a 100% comparison...


So it's not a comparison. To compare noise you need to equalise magnification. That means uprezzing the 5D file to the same pixel dimensions as the 1D4 file before comparing crops at 100%.
Feb 02, 2010 at 08:58 PM
David Manning
Offline
Details
p.1 #7Noise, Interpolation, and Comparison 101
M Vers wrote:
32067dlm wroteIf you said right, you'd be wrong!!
Hmm, what do the results look like if run through DPP? Also, what were the in camera settings? The result seems uncharacteristic for both the MKIV, at least from what I have seen so far, and the 5D.


The images weren't captured for comparison. I was simply wanting to begin setting LR3 import settings to automatically set chroma noise reduction. I don't use DPP, and I didn't intend a comparison, so I haven't installed DPP. I will say, I would use color files up to around ISO 8000 and Black and Whites up to ISO 25,600. This is all from 24 hours of ownership. I've got lots more learning.

1DIV (ISO 3200, 1/2500s, f/6.3)
5D (ISO 3200, 1/3200s, f/5.6)
Feb 02, 2010 at 08:58 PM
M Vers
Offline
Details
p.1 #8Noise, Interpolation, and Comparison 101
brainiac wrote:
32067dlm wroteThis is a 100% comparison...
So it's not a comparison. To compare noise you need to equalise magnification. That means uprezzing the 5D file to the same pixel dimensions as the 1D4 file before comparing crops at 100%.


Not only that, but my 5D shows much more chroma noise than what is posted above.
Feb 02, 2010 at 08:59 PM
David Manning
Offline
Details
p.1 #9Noise, Interpolation, and Comparison 101
brainiac wrote:
32067dlm wroteThis is a 100% comparison...
So it's not a comparison. To compare noise you need to equalise magnification. That means uprezzing the 5D file to the same pixel dimensions as the 1D4 file before comparing crops at 100%.


I understand your position. This is in no way a controlled "comparison", just an observation about per pixel noise. Frankly, the difference in size isn't that significant.
Feb 02, 2010 at 09:00 PM
David Manning
Offline
Details
p.1 #10Noise, Interpolation, and Comparison 101
M Vers wrote:
brainiac wrote
32067dlm wroteThis is a 100% comparison...
So it's not a comparison. To compare noise you need to equalise magnification. That means uprezzing the 5D file to the same pixel dimensions as the 1D4 file before comparing crops at 100%.
Not only that, but my 5D shows much more chroma noise than what is posted above.


Would you like the DNGs?
Feb 02, 2010 at 09:01 PM
M Vers
Offline
Details
p.1 #11Noise, Interpolation, and Comparison 101
32067dlm wrote:
Would you like the DNGs?


TBH I don't need them, I own the 5D and know what it is and isn't capable of. By now I'm sure everyone knows how much RAW conversion methods matter, specifically with new bodies. If you've got DPP and wouldn't mind shooting the same 'test', I'm sure a few of us would be curious to see the results

EDIT: making sure all NR is off (in camera and out) and the files are equalized as well

Edited on Feb 02, 2010 at 09:08 PM · View previous versions
Feb 02, 2010 at 09:02 PM
wickerprints
Offline
Details
p.1 #12Noise, Interpolation, and Comparison 101
32067dlm wrote:
I compared my 1D Mark IV to my original 5D at ISO 3200. This is a 100% comparison from Lightroom 3 beta with no noise reduction. Which camera is on which side?

David


If you are comparing both cameras at 100% crop, then (1) the result is unsurprising (2) your metric for comparing noise across cameras is incorrect.

The 1D4 is a 16MP APS-H camera. The 5D is a 12.8MP 35mm camera. This means:

1D4 = 30575 photosites / mm^2.
5D = 14722 photosites / mm^2.

That is over twice the photosite count per unit area, or pixel density. Thus you cannot compare the per-pixel noise and expect that they will be comparable. This is the same error that DPReview makes.

The only fair evaluation of noise is to look at noise per unit area of sensor. [EDIT: italicized statement is incorrect, see following posts. -w]

The below should be standard reading for anybody who cares to compare noise across different camera bodies.

http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/index.html

Edited on Feb 02, 2010 at 09:17 PM · View previous versions
Feb 02, 2010 at 09:06 PM
David Manning
Offline
Details
p.1 #13Noise, Interpolation, and Comparison 101
M Vers wrote:
32067dlm wroteWould you like the DNGs?
TBH I don't need them, I own the 5D and know what it is and isn't capable of.


Honestly, I'm surprised. I even went back to LR to see if I had forgot to turn of the noise control. I will say the chroma noise reduction in LR3beta is outstanding.
Feb 02, 2010 at 09:06 PM
wickerprints
Offline
Details
p.1 #14Noise, Interpolation, and Comparison 101
brainiac wrote:
32067dlm wroteThis is a 100% comparison...
So it's not a comparison. To compare noise you need to equalise magnification. That means uprezzing the 5D file to the same pixel dimensions as the 1D4 file before comparing crops at 100%.


+1. Again, brainiac says in a few sentences what I struggle to express in a whole paragraph.
Feb 02, 2010 at 09:07 PM
brainiac
Offline
[X]
Details
p.1 #15Noise, Interpolation, and Comparison 101
32067dlm wrote:
brainiac wrote
32067dlm wroteThis is a 100% comparison...
So it's not a comparison. To compare noise you need to equalise magnification. That means uprezzing the 5D file to the same pixel dimensions as the 1D4 file before comparing crops at 100%.
I understand your position. This is in no way a controlled "comparison", just an observation about per pixel noise. Frankly, the difference in size isn't that significant.


The difference is about 20%. You have given a 20% advantage to the 5D. This isn't about controlled shooting, it's about looking at your results without a major bias. Comparison of noise MUST be done by examining equivalent fractions of the frame. You haven't done that here, so your uploaded samples are very misleading. You can't just casually wave this away. If you took away 20% of one lens's sharpness in a lens test that would be very wrong, and it would be shameful to upload it and say it was in any way fair or informative. Casualness about magnification is a major problem in camera assessment here and on other sites. Do it right. Uprez the 5D file to 1D4 dimensions and then show 100% crops.
Feb 02, 2010 at 09:10 PM
David Manning
Offline
Details
p.1 #16Noise, Interpolation, and Comparison 101
wickerprints wrote:
brainiac wrote
32067dlm wroteThis is a 100% comparison...
So it's not a comparison. To compare noise you need to equalise magnification. That means uprezzing the 5D file to the same pixel dimensions as the 1D4 file before comparing crops at 100%.
+1. Again, brainiac says in a few sentences what I struggle to express in a whole paragraph.


Okay, look at the width of the black between the patches. There's not much difference in size. Again, just to be clear, I understand what you all are saying, and I agree. You are, however, seeing what I see.
Feb 02, 2010 at 09:13 PM
brainiac
Offline
[X]
Details
p.1 #17Noise, Interpolation, and Comparison 101
wickerprints wrote:
That is over twice the photosite count per unit area, or pixel density. Thus you cannot compare the per-pixel noise and expect that they will be comparable. This is the same error that DPReview makes. The only fair evaluation of noise is to look at noise per unit area of sensor.


NO! DO NOT COMPARE NOISE PER UNIT AREA! Compare noise per image. Comparing noise per unit area will tell you how the 5D would compare against a FULL FRAME 1D4. Nobody owns one of those. It's much simpler than that. You need to compare equal fractions of the image. To do that, just uprez one image to the dimensions of the other, and compare at 100%. Job done.
Feb 02, 2010 at 09:14 PM
wickerprints
Offline
Details
p.1 #18Noise, Interpolation, and Comparison 101
brainiac wrote:
wickerprints wroteThat is over twice the photosite count per unit area, or pixel density. Thus you cannot compare the per-pixel noise and expect that they will be comparable. This is the same error that DPReview makes. The only fair evaluation of noise is to look at noise per unit area of sensor.
NO! DO NOT COMPARE NOISE PER UNIT AREA! Compare noise per image. Comparing noise per unit area will tell you how the 5D would compare against a FULL FRAME 1D4. Nobody owns one of those. It's much simpler than that. You need to compare equal fractions of the image. To do that, just uprez one image to the dimensions of the other, and compare at 100%. Job done.


Yes, I misspoke when I wrote that. Thanks for pointing that out. See? It's pretty easy to get confused
Feb 02, 2010 at 09:15 PM
brainiac
Offline
[X]
Details
p.1 #19Noise, Interpolation, and Comparison 101
32067dlm wrote:
Okay, look at the width of the black between the patches. There's not much difference in size.


The width of the black is irrelevant. That depends on lens focal length and whether it is adjusted for the crop angle of view. But here we are trying to look at the sensor's performance, so the width of the black stripe doesn't matter, UNLESS you ensured that the angles of view were identical. IFF you ensured that the angles of view and distance to subject were identical, then the width of the strap would count as a guide to magnification. But all this is irrelevant. Just uprez the 5D file to 1D4 dimensions and show at 100%. Job done.
Feb 02, 2010 at 09:18 PM
brainiac
Offline
[X]
Details
p.1 #20Noise, Interpolation, and Comparison 101
wickerprints wrote:
brainiac wrote
wickerprints wroteThat is over twice the photosite count per unit area, or pixel density. Thus you cannot compare the per-pixel noise and expect that they will be comparable. This is the same error that DPReview makes. The only fair evaluation of noise is to look at noise per unit area of sensor.
NO! DO NOT COMPARE NOISE PER UNIT AREA! Compare noise per image. Comparing noise per unit area will tell you how the 5D would compare against a FULL FRAME 1D4. Nobody owns one of those. It's much simpler than that. You need to compare equal fractions of the image. To do that, just uprez one image to the dimensions of the other, and compare at 100%. Job done.
Yes, I misspoke when I wrote that. Thanks for pointing that out. See? It's pretty easy to get confused


It is easy to get confused. But the rule for avoiding bent comparisons is simple: to compare any two cameras, uprez the lower rez one to the dimensions of the higher, and compare at 100%. If you're looking for detail it will help if you have used the correct focal lengths to ensure that the pictures have identical composition and bounds. That's it. Pretty much any other method is highly susceptible to error.
Feb 02, 2010 at 09:28 PM
FM ForumsCanon ForumJoin Upload & Sell
  
 

You are not logged inLog inRegister