Understanding dynamic range of digital back

[nightfire]

While playing around with my recently purchased P45+, I took some pictures of a white wall to learn more the exposure characteristics of the back.

First, I exposed to a standard light meter reading of the wall (18% gray). The resulting histogram peak was in the center of the range displayed. So far, so good.

Next, I added exposure until the histogram clipped to the right, and decreased exposure until it clipped to the left. The result was +2 stops and -5 stops, which would yield a dynamic range of 7 stops.

However, the P45+ specs list a dynamic range of 12 stops, so obviously there must be something wrong with my understanding of dynamic range. Can anyone explain the difference to me?

[Graham Mitchell]

Does the back apply a contrast curve? If so, have you tried it with a linear setting?

[Doug Peterson]

Go another few stops each direction. Go into Capture One 5.2. Use highlight/shadow recovery or exposure adjustment (or switch to linear curve and play around with curves).

If the wall had texture to start with (i.e. not a solid glossy texture-less wall) and still has recoverable texture than it was within the DR of the system.

[bjanes]

While playing around with my recently purchased P45+, I took some pictures of a white wall to learn more the exposure characteristics of the back.

First, I exposed to a standard light meter reading of the wall (18% gray). The resulting histogram peak was in the center of the range displayed. So far, so good.

Next, I added exposure until the histogram clipped to the right, and decreased exposure until it clipped to the left. The result was +2 stops and -5 stops, which would yield a dynamic range of 7 stops.

Taking exposures of a white wall is a standard method of determining the characteristics of a sensor. You should use a lens with a relatively long focal length, defocus the lens, and use only the centrtal portion of the field to avoid light falloff. Roger Clark shows how to do this. You then need to use a program which looks at the raw data. Even with no exposure (e.g. lens cap on, high shutter speed, small aperture), you should still get a reading, and this is a measure of the read noise of the sensor. As previously pointed out, the clipping you saw is likely from the tone curve applied to the raw data. Also, the Histogram of Photoshop and many other programs does not have enough resolution to show the values at the low end of the DR.

Here is a histogram of a Nikon D3 with the lens cap on (the free ware program Iris was used). Nikons clip the read noise at zero and show only half of the bell shaped curve. Canons add an offset to prevent such clipping, and I don't know what Phase One does.

The engineering definition of DR is maximum signal/read noise. For practical photography, you might want to set the noise floor higher. The DR is limited by noise, not by clipping.

Bill

[nightfire]

Thanks for the hints. I didn't consider the recovery sliders in my first test since my other lesser cameras never produced much useful data beyond the clipping region anyway, so I ended up using the recovery options less and less. Revisiting yesterday's P45+ wall images again, however, I was surprised at how much detail is recoverable compared to my former DSLR. In the end, I was able to recover texture in a +4 to -8 stop range, which would correspond to the 12 stops listed in the specs. The most extreme exposure on each side looks very ugly though. For determing exposure in the field, I'll rather stick to a +3 / -7 stop range since the extremes still show aesthetically acceptable detail here.

[bjanes]

Thanks for the hints. I didn't consider the recovery sliders in my first test since my other lesser cameras never produced much useful data beyond the clipping region anyway, so I ended up using the recovery options less and less. Revisiting yesterday's P45+ wall images again, however, I was surprised at how much detail is recoverable compared to my former DSLR. In the end, I was able to recover texture in a +4 to -8 stop range, which would correspond to the 12 stops listed in the specs. The most extreme exposure on each side looks very ugly though. For determing exposure in the field, I'll rather stick to a +3 / -7 stop range since the extremes still show aesthetically acceptable detail here.

I don't know what lesser dSLR you were using, but the better Nikons and Canons do allow considerable highlight recovery. A good example of recovery with a Canon is shown in Jeff Schewe's Real World ACR book with his restoration of an image of the Niagra falls.

However, use of recovery in the determination of dynamic range is cheating since you are dealing with clipped channels. Because of white balance, the channels in a Bayer array sensor do not all clip at the same time with increasing exposures, and the green channel usually clips first, often followed by blue and then red. With over exposure and clipping of the green channel, recovery works by reconstructing the lost green values from the non-clipped channels. If all channels are clipped, recovery will not work. I doubt that recovery is any more successful the Phase 1 than with a high-end Canon or Nikon camera.

Another source for scientific testing of dynamic range is the DXO site. They demonstrate little difference in dynamic range between the phase 1 P 65+and the Nikon D3X. they do not use recovery.

Regards,

Bill

If you want accurate data, you really should not clip channels.the dynamic range tests on DPReview are a bit muddled, but they do discuss this limitation.

[ondebanks]

I thoroughly echo bjanes' recommendation to try a "Roger Clark" style analysis of your back. I did it for mine, and it was bang on the money for readnoise and DR. It is a very satisfying and educational test to take on.

BTW, one other reason why your original test underestimated DR is that the histogram has a certain width. If you stopped exposing when the right edge of the histogram clipped at the bright end and the left edge clipped at the dark end, you were not testing the same pixels. It would be more accurate to go until the peak of the histogram clipped at both ends, but that is harder to judge, and anyway it is not a true measure of DR (as bjanes said).