How to determine the dynamic range of a camera - my experiment

[Josh66]

The data determined by my experiment on the dynamic range of the camera allowed me to put together a characteristics curve, shown below:

View attachment 38380

As discussed in Light, Science and Magic, 4th edition, pg. 244-253, an ideal characteristics curve would be linear, with each change in aperture having a linear increase or decrease in the density of the grey. Actual sensor performance is different. I therefore was interested in how my camera would behave, and what latitude I had in terms of the dynamic range. The above graph shows the behaviour of my camera in transforming an 18% grey card to values on the resulting image.

I welcome comments from more experienced photographers as to the interpretation of the curve as seen above.

I've read that book (not sure which edition right now, but what you said sounds familiar) - if I remember correctly, when the author mentioned the "ideal, linear" curve - he also mentioned that we actually prefer the 'flawed' 's-shaped' curve. The ideal curve was only ideal as far as it applies to making a 'perfect' sensor. Correct me if I'm wrong...

edit
A linear curve would make for a muddy, low contrast image, and we like contrast (at least a little contrast) - or something like that...

 

[pgriz]

The data determined by my experiment on the dynamic range of the camera allowed me to put together a characteristics curve, shown below:

View attachment 38380

As discussed in Light, Science and Magic, 4th edition, pg. 244-253, an ideal characteristics curve would be linear, with each change in aperture having a linear increase or decrease in the density of the grey. Actual sensor performance is different. I therefore was interested in how my camera would behave, and what latitude I had in terms of the dynamic range. The above graph shows the behaviour of my camera in transforming an 18% grey card to values on the resulting image.

I welcome comments from more experienced photographers as to the interpretation of the curve as seen above.

I've read that book (not sure which edition right now, but what you said sounds familiar) - if I remember correctly, when the author mentioned the "ideal, linear" curve - he also mentioned that we actually prefer the 'flawed' 's-shaped' curve. The ideal curve was only ideal as far as it applies to making a 'perfect' sensor. Correct me if I'm wrong...

edit
A linear curve would make for a muddy, low contrast image, and we like contrast (at least a little contrast) - or something like that...

Yeah, you're right. On Pg. 254 of the 4th edition, the authors mention that while the RAW image captures sensor response being more linear, the RAW converters recreate the "shoulder" s-curves to match the behaviour we "expect". Unfortunately, each RAW converter does it a little differently, and the authors seem to favour the idea of converting the RAW to DNG format, mainly because the DNG format processing is open while the RAW processing is proprietary.

So based on that, what I was measuring was not the sensor response, but Canon's RAW conversion of the sensor response. In other words, if I use a different RAW converter, I may end up with a different curve.
Hmmm. That seems to reflect the earlier post by HelenB. So to find out what my dynamic range is, I have to also consider the RAW converter being used. UGH. This is getting complicated.

 

[Garbz]

Unfortunately, each RAW converter does it a little differently, and the authors seem to favour the idea of converting the RAW to DNG format, mainly because the DNG format processing is open while the RAW processing is proprietary.

That makes no sense. There's nothing about the proprietary vs open aspects of DNG that effect the result. To convert to DNG you still need to apply a calibration curve and colour profile. May as well convert to TIFF.

 

[pgriz]

I am no expert on this, as I'm trying to understand what is actually going on "under the hood", so to speak, but it appears that the arguement for using DNG is that (according to my understanding of the authors in the book I referenced) the DNG format is the same as RAW but without the proprietary coding. If, as you say, to convert to DNG you still need to apply a calibration curve and profile, then the questions becomes which calibration curve? I will make the assumption that each RAW converter has a different curve, and it would interesting to understand how that curve is developed. I have noticed that using the same RAW files as a starting point, and processing it with different brightness settings gave me more detail in both the highlight and shadow areas. So there appears to be more information in the RAW file, some of which is discarded when being converted to a format we perceive as the image. Therefore, in principle, I should be able to produce three versions of the same RAW file, and merge them using some form of HDR to get access to the true dynamic range of the sensor. Does that approach make sense to you?

 

[Parker219]

Some pics in this thread would be nice!

 

[Garbz]

it would interesting to understand how that curve is developed. I have noticed that using the same RAW files as a starting point, and processing it with different brightness settings gave me more detail in both the highlight and shadow areas. So there appears to be more information in the RAW file, some of which is discarded when being converted to a format we perceive as the image. Therefore, in principle, I should be able to produce three versions of the same RAW file, and merge them using some form of HDR to get access to the true dynamic range of the sensor. Does that approach make sense to you?

What you're describing is exactly the workaround many people used for HDR programs before they natively supported opening RAW files. In reality you just need to compress the dynamic range of the image before loading it into the RAW program. A 16bit file has enough dynamic range for all the data, you just need to ensure when you convert from RAW to TIFF you don't clip any highlights or shadows and let the HDR program take care of the rest.

The argument for DNG doesn't make any sense because it doesn't solve the fundamental program of the curve and colour calibration. If you use a program to convert the RAW to DNG then you're at the mercy of that program's settings. So why not just use that program to edit your RAW file anyway? If your camera natively performs DNG you get even less choice.

As to how the curves are determined. It's typically either something representing Gamma 2.2 or the L* correction curve. Actual contrast settings then vary between manufacturer and so does actual colour space conversion. Adobe Labs provide a nice tool to allow you to customise your colour curve if you don't like the way Adobe Standard responds, and some manufacturers allow you to make custom DNG colour profiles (which is what all Adobe programs use to convert RAWs) based on calibration charts. It's all very unscientific.

 

[Helen B]

The conversion from the camera raw to a DNG file does not usually affect the data: no 'calibration curve' or profile is applied to the data, the numbers remain the same. You can't apply a camera profile (other than a very simple matrix one that only defines the pure red, green and blue points) to raw data until it has been debayered, because the influence of the profile on a pixel is determined by the three-dimensional location of the pixel's colour values, therefore the other two values have to be computed first.

 

[Garbz]

The conversion from the camera raw to a DNG file does not usually affect the data: no 'calibration curve' or profile is applied to the data, the numbers remain the same.

Helen I was under the impression that while the numbers remain the metadata in DNG allows for all adjustments and changes including any information needed to render the file to be embedded. Certainly at the very least all RAW adjustments allowed in the ACR are embedded directly into the file INCLUDING a custom camera colour profile.

Without this information how could the DNG file be decoded in the future if it's a bit for bit copy of the proprietary RAW information? My thoughts were that the RAW data is converted to an open and standard data format (thus necessitating some interpretation of the proprietary data). Also the DNG spec allows for embedding the original RAW file which they note will approximately double the file size. The existence of this feature makes me think it's not a straight out copy of the data.