Ad
Ad
Ad
Pages: [1] 2 »   Bottom of Page
Print
Author Topic: Show me high ISO noise  (Read 2098 times)
jerome_m
Sr. Member
****
Offline Offline

Posts: 555


« on: January 25, 2014, 06:47:18 AM »
ReplyReply

I tried to devise a simple test to show comparison images taken with various cameras. The idea behind the test is to use subjects which are widely available so that other people can redo the same test with a different camera, even if they live somewhere else on the planet.

Since the object is to test high iso noise, we need the following objects:
-some colour reference object, I chose to use the widely available X-rite chart.
-some objects with fine structures and muted colours. I chose to use banknotes, which have the perfect combination of features and are widely available. But please keep in mind that we do not want to counterfeit money and that photoshop includes a system that will jump at you if you crop a complete banknote image, so leave some part of the notes out when you crop the picture.

The reference objects are attached to a wall (just make sure they lay really flat), the camera has a standard 80mm lens, distance is 2.5 m, the scene is lit with daylight (a nearby window). I suppose a studio flash could also be used. The aperture is constant at f/11 (compromise between depth of field and diffraction) and I vary the shutter time (vary the flash output if you use flash). If I were to vary the aperture, lens sharpness would vary.

Exposure is measured with an incident light meter. If you use the built-in meter, compensate about +1ev, since you are mainly taking a picture of a white wall or use a grey card.

Attached is the resulting complete picture, just scaled down.
Logged
ErikKaffehr
Sr. Member
****
Offline Offline

Posts: 7421


WWW
« Reply #1 on: January 25, 2014, 06:54:20 AM »
ReplyReply

Hi,

A decent proposal, count on me! I'll post images from P45+, Alpha 900 and Alpha 99 in a few days.

Best regards
Erik


I tried to devise a simple test to show comparison images taken with various cameras. The idea behind the test is to use subjects which are widely available so that other people can redo the same test with a different camera, even if they live somewhere else on the planet.

Since the object is to test high iso noise, we need the following objects:
-some colour reference object, I chose to use the widely available X-rite chart.
-some objects with fine structures and muted colours. I chose to use banknotes, which have the perfect combination of features and are widely available. But please keep in mind that we do not want to counterfeit money and that photoshop includes a system that will jump at you if you crop a complete banknote image, so leave some part of the notes out when you crop the picture.

The reference objects are attached to a wall (just make sure they lay really flat), the camera has a standard 80mm lens, distance is 2.5 m, the scene is lit with daylight (a nearby window). I suppose a studio flash could also be used. The aperture is constant at f/11 (compromise between depth of field and diffraction) and I vary the shutter time (vary the flash output if you use flash). If I were to vary the aperture, lens sharpness would vary.

Exposure is measured with an incident light meter. If you use the built-in meter, compensate about +1ev, since you are mainly taking a picture of a white wall or use a grey card.

Attached is the resulting complete picture, just scaled down.
Logged

jerome_m
Sr. Member
****
Offline Offline

Posts: 555


« Reply #2 on: January 25, 2014, 06:57:35 AM »
ReplyReply

How to interpret the results? The idea is that people do the same test with different cameras so that we can compare the results for visible grain (in the colour chart) and detail or lack thereof in the banknotes. We match pictures which gives the same visual results and check the respective iso value. If one camera gives comparable pictures at iso800 as the other one at iso1600, the second is 1 ev better, like in a real life situation.

The camera I choose to present as "worst case scenario" is an old Hasselblad H3D-31. Its crop factor is 1x1.3 and its resolution is 6496x4872 pixels. Images are treated with Phocus with everything as default (just the white balanced was adjusted on the X-rite chart central gray patch).

These are the results at iso 100, 200 and 400:

Logged
jerome_m
Sr. Member
****
Offline Offline

Posts: 555


« Reply #3 on: January 25, 2014, 07:03:48 AM »
ReplyReply

And these are the results at iso 800 and 1600. Now, I would be interested with results taken with the 40 mpix sensors (H4D-40 with Phocus and P40 with capture one) and obviously the New IQ250. All these cameras have a sensor with the same dimensions but they do not have the same resolution. If we want to compare the noise as it would be seen on prints, the images should just be resized to 6496x4872 pixels before the crop (but you can also post the full resolution).

Logged
jerome_m
Sr. Member
****
Offline Offline

Posts: 555


« Reply #4 on: January 25, 2014, 07:09:55 AM »
ReplyReply

I'll post images from P45+, Alpha 900 and Alpha 99 in a few days.

You are welcome to try, but these three cameras are very different than the 1.3 crop MF cameras, so we would need a way to compensate for different resolution, field of view and sensor size. And, BTW, what noise reduction software do you plan on using? Here I am using the manufacturer's default, but it is much better at hiding noise than what Sony delivers for the A900, for example.
Logged
ErikKaffehr
Sr. Member
****
Offline Offline

Posts: 7421


WWW
« Reply #5 on: January 25, 2014, 07:16:28 AM »
ReplyReply

Hi,

OK, if you are not interested, I don't need to send images, but I still feel it is a good initiative.

Regarding noise reduction I would probably use noise reduction set to zero in ACR or even use RawDigger. I want something that shows sensor characteristics.

Best regards
Erik

You are welcome to try, but these three cameras are very different than the 1.3 crop MF cameras, so we would need a way to compensate for different resolution, field of view and sensor size. And, BTW, what noise reduction software do you plan on using? Here I am using the manufacturer's default, but it is much better at hiding noise than what Sony delivers for the A900, for example.
Logged

jerome_m
Sr. Member
****
Offline Offline

Posts: 555


« Reply #6 on: January 25, 2014, 07:58:38 AM »
ReplyReply

Regarding noise reduction I would probably use noise reduction set to zero in ACR or even use RawDigger. I want something that shows sensor characteristics.

I was afraid you would suggest that. While this is a good approach when considered in isolation, it is very remote from common photography practice. Most people use more standard software like Adobe, Capture one or Phocus and these include noise reduction by default even when you set it to zero. Besides, there is growing concern that even the raw data includes some noise reduction of sort. Just testing the raw data wold just reward the manufacturers who are dishonest and add hidden noise reduction.

I thus think it is fairer and more interesting to display images with noise reduction as the manufacturers included it. This is the reason for the banknotes BTW: to be able to see the compromise between smoothing of noise and destruction of fine details at higher iso.
Logged
jerome_m
Sr. Member
****
Offline Offline

Posts: 555


« Reply #7 on: January 25, 2014, 08:06:49 AM »
ReplyReply

For everybody's amusement and since the setup is still on, I redid the test with a D800 (not "e", the standard one). The lens used is a 60mm micro-Nikkor at f/5.6, the result is directly comparable to the H3D-31. These are extracted from the jpegs, so as to test Nikon's built-in noise reduction. I also have the raw files, if someone is interested.

A word of caution: iso 400 appears to be minimally blurred by shutter movement. This is the best I can do amongst a few tries on a decent tripod, so shutter problems are not a "Sony A7r only" feature…

These are iso 100, 200, 400 and 800. The effect or noise reduction on fine detail is particularly apparent on the blue eagle of the dollar bill.

Logged
jerome_m
Sr. Member
****
Offline Offline

Posts: 555


« Reply #8 on: January 25, 2014, 08:08:58 AM »
ReplyReply

And these are iso 1600, 3200, 6400 and 12800.
Logged
jerome_m
Sr. Member
****
Offline Offline

Posts: 555


« Reply #9 on: January 25, 2014, 08:19:07 AM »
ReplyReply

A direct comparison of the H3D-31 with the D800 shows the different approaches taken by the manufacturers. While the H3D-31 keeps fine detail visible even at its max sensitivity, the D800 delivers smoother X-rite colour patches but fine detail in the banknotes becomes invisible very quickly. The results are otherwise relatively close at the same sensitivity, except iso 1600 on the H3D-31. But the H3D-31 does not really expose at iso 1600, it just pushes the iso 800 digital data. It is an "extended sensitivity" on that particular back, just as iso 12800 on the D800.
Logged
eronald
Sr. Member
****
Offline Offline

Posts: 4034



« Reply #10 on: January 25, 2014, 08:42:20 AM »
ReplyReply

Noise depends on the light because raw channels are not balanced.
You need to specify what light you use.
Logged
Paul2660
Sr. Member
****
Offline Offline

Posts: 1827


WWW
« Reply #11 on: January 25, 2014, 08:50:54 AM »
ReplyReply

Thanks for the test.  I have always assumed that high iso jpgs taken in camera are pretty worthless, unless they are for web publishing.  I prefer not to any camera's in house processing as you will always be limited by the processor of the camera.  

I find from shooting the D800 at around 1600 to 2000 that I can still get a very good file as long as I stay in raw.  As I understand LR 5.3 a small bit of sharpening is applied from the initial import of the file but noise reduction is neutral.  Capture One on the other hand dials in both a preset amount of sharpening and noise reduction based on the camera file it's working with.  I have always found that Capture One's noise reduction especially on high iso files is a bit harsh and can quickly get to the watercolor look or other strange effects, but Capture One also offers some excellent sliders to give very precise control for the end user.

For my work, I find the high iso range of the D800 in 1600 to 3200 range acceptable.  Works great in medium light when I am working birds or other wildlife in DX mode and the rare times I am working in low light without a tripod.

I fully expect to see an improvement with the IQ250 over the Sony 36MP chip as the pixels are larger, and thus should have the ability to handle lower amounts of light better.  Not to mention the sensor has a considerably larger surface area.

Paul Caldwell
Logged

Paul Caldwell
Little Rock, Arkansas U.S.
Photography > http://photosofarkansas.com
Blog> http://paulcaldwellphotography.com
jerome_m
Sr. Member
****
Offline Offline

Posts: 555


« Reply #12 on: January 25, 2014, 08:57:22 AM »
ReplyReply

Noise depends on the light because raw channels are not balanced.
You need to specify what light you use.

But I did (in the first post). I used daylight from a nearby window. As flash has the same colour temperature, I suppose it would give the same results. Incandescent light would indeed give different results, probably noticeably worse (there would be more noise in the blue channel).
Logged
ErikKaffehr
Sr. Member
****
Offline Offline

Posts: 7421


WWW
« Reply #13 on: January 25, 2014, 09:04:00 AM »
ReplyReply

Hi,

I would suggest that raw images need to be included.

Best regards
Erik

But I did (in the first post). I used daylight from a nearby window. As flash has the same colour temperature, I suppose it would give the same results. Incandescent light would indeed give different results, probably noticeably worse (there would be more noise in the blue channel).
Logged

jerome_m
Sr. Member
****
Offline Offline

Posts: 555


« Reply #14 on: January 25, 2014, 09:05:25 AM »
ReplyReply

Thanks for the test.  I have always assumed that high iso jpgs taken in camera are pretty worthless, unless they are for web publishing.  I prefer not to any camera's in house processing as you will always be limited by the processor of the camera.  

I find from shooting the D800 at around 1600 to 2000 that I can still get a very good file as long as I stay in raw.  As I understand LR 5.3 a small bit of sharpening is applied from the initial import of the file but noise reduction is neutral.  Capture One on the other hand dials in both a preset amount of sharpening and noise reduction based on the camera file it's working with.  I have always found that Capture One's noise reduction especially on high iso files is a bit harsh and can quickly get to the watercolor look or other strange effects, but Capture One also offers some excellent sliders to give very precise control for the end user.

Nikon jpegs should just be seen as a reference point. But I agree that with raw capture and an external noise reduction software one could probably improve the results, even if Nikon's processing is already quite good.


Quote
For my work, I find the high iso range of the D800 in 1600 to 3200 range acceptable.  Works great in medium light when I am working birds or other wildlife in DX mode and the rare times I am working in low light without a tripod.

I fully expect to see an improvement with the IQ250 over the Sony 36MP chip as the pixels are larger, and thus should have the ability to handle lower amounts of light better.  Not to mention the sensor has a considerably larger surface area.


The end results will depend on which size you want to print and the IQ250, as well as the venerable H3D-31 presented here win on sensor capture size. But less us not forget the real question. The real question is: how many stops does one win between the new IQ250 and, say, an H5D-40 / P40+ / IQ140?
Logged
jerome_m
Sr. Member
****
Offline Offline

Posts: 555


« Reply #15 on: January 25, 2014, 09:08:33 AM »
ReplyReply

I would suggest that raw images need to be included.

I can send you the raw images, but would suggest that you start another thread if you want to post the results of your conversions.
Logged
Vladimirovich
Sr. Member
****
Offline Offline

Posts: 1320


« Reply #16 on: January 25, 2014, 10:25:03 AM »
ReplyReply

I can send you the raw images, but would suggest that you start another thread if you want to post the results of your conversions.
noise can be analyzed w/o conversions though for as long as tools like rawdigger can read your raws... or are you in fact testing NR in raw converters ?
Logged
jerome_m
Sr. Member
****
Offline Offline

Posts: 555


« Reply #17 on: January 25, 2014, 10:38:11 AM »
ReplyReply

noise can be analyzed w/o conversions though for as long as tools like rawdigger can read your raws... or are you in fact testing NR in raw converters ?

As a photographer, I am mainly interested in the end results and that is: prints. Large prints, or I don't need a MF. Before printing, noise reduction is a standard operation when using large iso values.

If people want to test raw digger, that is fine with me, I even offered to deliver the files. But the test I am interest in and I ask for is: take your best camera, hang banknotes and a colour chart on a wall, do your best with whatever software you fancy and show me the results. I am particularly interested in what the following cameras can do:
Hasselblad H5D-40 with Phocus
Pentax 645D
IQ140 or P40+ with Capture One
IQ250 with whatever you want.

What is so difficult to understand?
Logged
digitaldog
Sr. Member
****
Offline Offline

Posts: 9100



WWW
« Reply #18 on: January 25, 2014, 10:54:31 AM »
ReplyReply

Might be useful to point out that the camera itself and how it deals with proper exposure for raw (ETTR) can play a role? IOW, a higher ISO can produce less noise than a lower one:

This is a Canon 5DMII.
Logged

Andrew Rodney
Author “Color Management for Photographers”
http://digitaldog.net/
jerome_m
Sr. Member
****
Offline Offline

Posts: 555


« Reply #19 on: January 25, 2014, 11:22:06 AM »
ReplyReply

A fair warning, which is why I advise to use an incident light meter.
Logged
Pages: [1] 2 »   Top of Page
Print
Jump to:  

Ad
Ad
Ad