Shutter speed and ISO correction


I would like to be able to compare images taken at different shutter speed and ISO.
Typically, one can use an equation such as this one to convert pixel values :
p′ = p( k^2/τK)
, where p is original pixel value, p’ is the correrted pixel value, k is aperture, τ is shutter speed and K is ISO.

But it does not work with the Sequoia as you can see. this is three images at the same ISO but at different shutter speed of 6 targets with known reflectance.

Could you explain how to apply the shutter speed and ISO correction?

Questions about Radiometric Calibration and data correction
Details of Irradiance List tag for Sunshine sensor in exif data of Sequoia
Details of Irradiance List tag for Sunshine sensor in exif data of Sequoia

@pk123, you result makes me believe that new pixel value is a function of reflectance target as well.

It does seem that relationship is linear but then again it doesn’t seem to have a zero intercept as one would expect looking at the equation.


@muzammil360, indeed the intercept is not 0, and what’s strange is that the intercept varies with the reflectance of the targets. Actually, the intercept correlates perfectly with the reflectance of each target.
So, this is not your usual sensor response…


Does anyone have an explanation for these results?


@pk123 and @muzammil360 does the formula above in terms of the difference in exposure value probably only apply for a linear camera response or at least for the portion of the camera response that is linear?

Recovered response from Sony F-717 camera, measurement from the MacBeth chart as ground truth (dots).


@seanmcleod, the info you are presenting relates to the relationship between reflectance and pixel value, not exposure. For the sequoia, the relationship between reflectance and pixel value is linear.


Actually the non-linear graphs of irradiance to pixel intensity only apply if the camera is applying a gamma curve as part of it’s conversion to jpeg etc. right?

From: Computer Vision: Algorithms and Applications

“Even more simply, you can just assume that the response is linear if they are RAW files and that the images have
a gamma = 2:2 non-linearity (plus clipping) applied to each RGB channel if they are JPEG images.”

I’m assuming the Sequoia doesn’t apply any gamma correction when writing out the multispectral tiffs right?

So ignoring other effects like vignetting the camera response should be pretty linear?


@seanmcleod, yes it is linear, see previous response.


Yep, what was confusing to me was that pretty much all the papers on the camera response function described it’s non-linear nature and virtually none of them explicitly stated that this only applies when the camera performs gamma correction etc. (ignoring vignetting etc.) and it doesn’t apply to RAW files which store the sensor values without applying gamma etc.

Maybe it’s obvious to practitioners in the art :wink:


All the answers should now be in the first application note about the CMOS sensor model
Don’t hesitate to post your findings with the CMOS sensor model here to further the discussion.