Small bomb from LL and DxO

That is a great read. I would definitely love to see the GF1/GH1/GH2 chart. That would probably answer most of our questions.

I would also love to see the graph for the new Sony A55 camera. It just has to look terrible against the other cameras when they throw away 30% of the light.
 
Interesting read. So there seems to be less of an issue with sensors that have gapless micro-lens design, such as 5D and 3Ds.

I wonder how they were able to measure the amount of gain that cameras apply to compensate for light loss.

Also, I see that LL has GH2 in for testing. Looking forward to their review ...
 
Last edited:
Wow. I am assuming that this only happens with lenses made by the manufacturer that communicate with the camera? A manual lens with no electronic communication should behave normally because the camera shouldn't have any data to compensate for the lower T-stop via ISO (as it won't know what lens or aperture it's dealing with.) One more reason to be using Nikkor lenses on your Canons for video :)
 
Ben_B said:
Wow. I am assuming that this only happens with lenses made by the manufacturer that communicate with the camera? A manual lens with no electronic communication should behave normally because the camera shouldn't have any data to compensate for the lower T-stop via ISO (as it won't know what lens or aperture it's dealing with.) One more reason to be using Nikkor lenses on your Canons for video :)
Click to expand...

Or Nikon lenses on your m4/3s cameras.
 
Luminous Landscape said:
This graph clearly shows that camera manufacturers “game the system” by increasing the ISO without the photographer's knowledge.
Click to expand...
These findings by DxOMark and Luminous Landscape cast a harsh light on one of the most archaic conventions used in modern cameras:

ISO is an obsolete standard for digital photography.

In the film era, ISO was a fundamental physical property of each film type and served as a natural starting point for determining proper exposure. As auto-exposure camera systems developed, shutter speed and aperture were treated as the primary adjustable exposure variables. Film ISO remained a fixed reference point that could only be manually adjusted by the user as exposure compensation.

In a digital camera, there is no physical recording medium that imposes a fixed ISO rating on exposure. ISO is nothing more than an arbitrary numerical scale used to adjust the video gain applied to the RGB levels generated by the image sensor. It is inherently stepless, and in many camcorders it is directly adjustable as an analog-style gain control.

In digital still photography, however, ISO is persistently treated as if it were a fixed reference point. In manual exposure mode, the user presets ISO in coarse 1/3rd-stop increments, while manually adjusting shutter speed and exposure as desired. For semi-automatic exposure modes, the user has a choice between manually setting either shutter speed or aperture, while the camera automatically adjusts the other for proper exposure.

What digital cameras should allow a photographer to do is manually set both shutter speed and aperture as desired. In auto-exposure mode, the camera should then adjust the video gain for proper exposure (though all three settings should continue to be user-adjustable in manual exposure mode). In digital recording, there is always an optimal gain setting that preserves detail in the brightest areas of the image. Unlike the artistic aspects of shutter speed and aperture selection, setting the proper video gain is noting more than a technical adjustment that makes best use of the digital recording format.

ISO is an obsolete concept that should be replaced by manual and/or automatic gain control.
 
lpowell said:
What digital cameras should allow a photographer to do is manually set both shutter speed and aperture as desired. In auto-exposure mode, the camera should then adjust the video gain for proper exposure (though all three settings should continue to be user-adjustable in manual exposure mode). In digital recording, there is always an optimal gain setting that preserves detail in the brightest areas of the image. Unlike the artistic aspects of shutter speed and aperture selection, setting the proper video gain is noting more than a technical adjustment that makes best use of the digital recording format.
]
Click to expand...

GH1 allows this AFAIK. Set video exposure mode to M and ISO to auto.
 
.very interesting.. it confirms a suspect i had on the stock 14-140 lens, that it is actually slower than stated.. when i did a couple of tests to compare stock lens with various FD lenses i have, i found that i always had to open up iris more on stock lens at same focal length and same lighting and shutter compared to other lenses..
try filming something with a FD 50mm lens @ say 5.6 .. adjust your shutter to get correct exposure. Shoot the same scene with stock lens @50mm focal and same shutter and you'll see that 5.6 is not enough...
 
Ben_B said:
Wow. I am assuming that this only happens with lenses made by the manufacturer that communicate with the camera? A manual lens with no electronic communication should behave normally because the camera shouldn't have any data to compensate for the lower T-stop via ISO (as it won't know what lens or aperture it's dealing with.) One more reason to be using Nikkor lenses on your Canons for video :)
Click to expand...

From what I can understand,
in this case, however, you trade some unknown exposure (light loss not gain compensated) in exchange of unknown less noise (by avoiding the gain compensation):
so, in order to have a correcly nominal exposed picture we have to manually "overexpose" it to counterbalance the disabled gain campensation due to manual lens,
and the amount of manual overexposure to apply is variable case-by-case and not a priori known.

Now, the really funny thing is that:
if we try to compensate by using longer shutter speed maybe we get blurred pictures as side effect,
if we try to compensate by increasing the ISO we get again more noise (the noise we wanted to avoid initially),
and if we try to compensate by widening the aperture (if possible) it will obviously change the DOF but it will change all the balls in the game again as side effect...! :)

The bottom line is that there is no win here:
- using manual lenses the "light loss at the sensor" is still not recoverable (including the decreased bokeh blur as side effect)
and the user must manually apply a variable unknown amount of overexposure to really get the intended correct exposure (with all the related side effects), or simply accept the variable unknown not correct exposure
- using the manufacturer electronic communicating lenses, on the other side, it will force the user to accept a variable unknown amount of noise added in-camera with the automatic gain curve applied.

Your choice... :(
 
Last edited:
More control is always better than less control.

In any case, I somehow get the feeling that older lenses will have T-stop values closer to their F-stop values, though this may be entirely unwarranted. I just figure the less things are mucked up with electronic mumb-jumbo the better.
 
Ben_B said:
I just figure the less things are mucked up with electronic mumb-jumbo the better.
Click to expand...

I agree, honestly everything's gotten a bit too confusing, and I'm a very technical person. I hate having to think about what the camera MIGHT be doing completely without my knowledge which will affect my image.

There's another side to gain that hasn't been discussed that much either, which is white balance. Because cameras have more green sensors than red and blue, that *native* white balance (ie no gain applied to any channel in specific) will be very green. All digital cameras then add a certain amount of gain to red and blue channels to compensate, the amounts depending on your white balance setting.
It gets more confusing when you realize that sometimes part of this gain can be analog sensor channel gain, and some gain can be applied digitally after it is captured. It's pretty hard to tell the difference.

As a stills shooter on the GH1, when I want the least noise and most dynamic range, I shoot in RAW UniWB (you can make a custom white balance that applies no gain to any channels, and gives a green image, by hugely over-exposing a light source as your "grey card.")
I did this so that the histograms on review I would get, would accurately describe whether or not highlights had really been clipped in the RAW file or simply post-white-balance-gain.
However, after running some tests, I was really surprised to notice that an identical image taken with a regular white balance had more shadow noise than a UniWB image of the exact same exposure, with only Lightroom applying regular white balance channel gains later. This makes me think that part of the GH1's white balance is an analog channel gain stage, which actually happens to be noisier than doing the same thing with digital gain later instead.

Additionally, my testing found that exposing to-the-right at a *HIGHER* ISO and then later reducing the exposure of the RAW in post, will result in less noise than matching that exposure in camera by using only a lower ISO setting (shutter and aperture staying the same). So in some cases, more gain means less noise. This has lead to a general rule-of-thumb for RAW shooting (I wouldn't apply this to video necessarily, or JPEG): when in a low-light situation where you can't open aperture or decrease shutter more, it's actually better to increase the ISO as high as you possibly can without clipping, than to use a lower ISO.

Really makes you think. A lot of the things we're told about cameras have additional details you have to take into account to realize the full picture, and in some cases the assumptions you make on them are totally incorrect.
 
Last edited:
Along the same lines of thinking as lpowell's post, I just wanted to contest something the author of the article wrote,

"1. When I select a specific ISO, I have a reason to do so. I find it unacceptable for the camera to change it without my knowledge."

In digital cameras, this will happen regardless of the transmission issue. Usually due to white balance. Camera will apply a massive, VARIABLE amount of channel gain - which is really the exact same mechanism as ISO gain - as soon as you apply a white balance that is not UniWB. You will never have full control of your gain unless you expect to shoot in UniWB all the time.
Honestly, ISO settings mean almost nothing at all. To the author: you're deluded if you think that on any digital camera you have the opportunity to directly affect ISO in a linear way.
Not to mention all the dishonesty in camera's published ISO values.
Then again, I've been reading that ISO is not necessarily objectivelly quantifiable anyway, it's just an arbitrary unit based on rules-of-thumb more than anything else, so perhaps we shouldn't fault them for not using the same measurement techniques that DxO does.

What I would really like to see is seperate R,G,B gain meters. Even if you keep the controls the same and keep ISO around, at least show us a numeric value of what the camera is doing. Better, add a mode that also allows us to control individual channel gain at both an analog and digital level.

I'd also like to see a reduction of the green pixel sites used, and more red and blues. The original explanation of it matching the human eye's sensitivity to green doesn't make sense when you consider that the camera ends up adding massive gain to red and blue to compensate anyway.
 
Last edited:
Ben_B said:
More control is always better than less control.
Click to expand...

in line of principle I agree with you,
but there are some problem to reach our intended "control" target, if we discover that we do not have ALL the required informations and their related managing variables at our disposal...
I understand that the T loss is a "system" variable (lens part + camera sensor part), and with a given camera by choosing good lens you can enhance only that part,
if the sensor is a good sensor maybe the noise added by the auto gain compensation is inside a safe zone for the majority of real world every day use (not very low light situations obviously) and in these majority of cases the approach of the camera manufacturers maybe is the most effective and simple for the user.

The weird issue is that no camera manufacturer told clearly this story AFAIK:
this is the more disappointing part... :(

Regarding the white balance,
personally I tend to keep it out here, if the discussion is aimed to serious stills shooters,
because WB is only related to jpeg development and it is possible to manage them in post if you have the RAW.
Obviously, who shoot mainly in jpeg (and for the movies too, because color correction is not painless)
this could be an additional variable indeed,
(but at least we were already aware of that...)
 
alphaomega said:
because WB is only related to jpeg development and it is possible to manage them in post if you have the RAW.
Click to expand...

If you'll read my posts you'll see that my tests have shown this isn't the whole story, and that WB actually does affect the RAW by adding more noise than necessary.
I think some of the colour channel gain is being "baked in" by analog gain rather than post-capture digital gain. Somehow this actually is noisier than digital gain later.
 
dasbin said:
I'd also like to see a reduction of the green pixel sites used, and more red and blues. The original explanation of it matching the human eye's sensitivity to green doesn't make sense when you consider that the camera ends up adding massive gain to red and blue to compensate anyway.
Click to expand...
The RGBG Bayer Filter doubling of green image sensors is a good approximation of the YCC conversion calculation of RGB luminance level:

Y = 30% R + 59% G + 11% B

If you summed the photon levels in a square RGBG array you'd get this:

Y = 25% R + 50% G + 25% B

Which is not a bad approximation of the actual calculation. If there weren't double the number of green pixels, the camera would have to multiply the sensor's green levels by a compensating factor of two, which would also double the amount of green pixel noise.
 
Is that luminance formula supposed to be post- or pre- white-balance? Or does it matter?

Does this mean that the current RGBG Bayer layout will result in reduced luminance noise, but increased color noise, as compared to a purely RGB layout?

And why does luminance matter at all if the image stays in the RGB realm the whole time?
 
dasbin said:
Is that luminance formula supposed to be post- or pre- white-balance? Or does it matter?

Does this mean that the current RGBG Bayer layout will result in reduced luminance noise, but increased color noise, as compared to a purely RGB layout?

And why does luminance matter at all if the image stays in the RGB realm the whole time?
Click to expand...
The RGBG Bayer Filter output is only available in RAW still images. For both JPEG and video files, 12-bit RGBG is white-balanced, converted into 8-bit YCC color space, and compressed into JPEG, MJPEG, or AVCHD files.

The complete set of YCC conversion calculations show how the RGB components are distributed:

Y = 30% R + 59% G + 11% B
Cb = 50% B - 33% G - 17% R
Cr = 50% R - 42% G - 8% B

(The minus signs are of no significance to this discussion.)

Since B and R pixel sensors each comprise 25% of the RGBG array, their photon levels must be multiplied by two to produce 50% of the Cb and Cr levels, respectively. However, noise from the other two colors is reduced by about the same factor, so total noise remains about the same. In addition, human vision is much less sensitive to chroma noise than luma noise, justifying the emphasis on reducing noise in the Y calculation.

The YCC distribution of colors in the Cb and Cr components is quite well balanced, as can be appreciated in this CbCr color slice at Y = 50% illumination (Cb - horizontal axis; Cr - vertical axis):

600px-YCbCr-CbCr_Y50.png
 
Last edited:
dasbin said:
If you'll read my posts you'll see that my tests have shown this isn't the whole story, and that WB actually does affect the RAW by adding more noise than necessary.
Click to expand...

Hello,
Sorry, I was writing and didn't see your previous post...

Let me understand clearly:
you say that you have found noticeable noise difference in the RAW output of the exact-same condition shot, taken only with different WB?
Honestly this sounds new and wondering to me...
Did you made this kind of test only with Panasonic or even with other cameras (Nikon, Canon...)?

BTW, LL covered the expose-to-right (ETR) trick as example here:
http://www.luminous-landscape.com/tutorials/expose-right.shtml
and I find interesting that in some cases you are still able to get better results performing ETR through higher ISO:
I can imagine that, in that particular situation, the advantage of quantization-shifts is bigger than the disadvantage
of the higher ISO noise introduced.
(However I think that you can still get even better results, in the same situation, performing ETR by normally increasing the exposure).
 
You must log in or register to reply here.
Back
Top