geonahta wrote:
Great thread and thanks to everyone.
Two points:
1) Noise in the underexposed shots might be affected by shutter speed. Anyone interested in doing a comparison under different lighting situations?
2) Anyone interested in doing the test 3 times while replacing the gray card with red/green/blue cards?
tayo wrote:
Ok, but without more information on how the histogram windows work, it might be that you are just extracting the information from the default profiles supplied with each RAW converter for that camera type (which are of course similar), and not testing the individual camera itself very much.
Can you elaborate more ? I don't understand the question. Thanks !
I really like the idea of using the Kodak gray step wedge as the basis for the test. But not in a single exposure. I did a preliminary test shooting the gray steps at variable shutter speeds. Then I open a file in ACR, do a white balance, and then look only at the middle gray step and the step next to it while I am making the exposure adjustment. Can I distinguish the middle step from those around it? That's the test to see if there is really detail in the highlights or details in the shadows.
My first test on the 5D seems to show 9 stops of dynamic range and that's conserative because the low shadow noise combined with some noise reduction will allow me to go lower.
The biggest advantage of the gray steps is that you can see if there are any weird artifacts appearing in the different gray levels when you need to make large exposure adjustments. There may be color shifts that you find unacceptable. So you may have to take that into account when you talk about dynamic range, it's not the maxiumum range that you want, it the maximum usable range.
Thet's some excellent, real world information in your summation sheet as it shows the 5D/1DsMkII to be the DR leaders, Leica aside and one or two more iterations and we should all be able to afford a 9/10 stop ranged sensor body
Pondria, here's a tidbit of eye science that relates to this discussion. I'm a physicist, who among other jobs, designs RGB (red green blue) LED backlights for LCDs (liquid crystal displays). LED (light emitting diode) intensity is measured in units called milli-candelas. These are eye-weighted units; one millicandela each of red, green, or blue, all appear to be equal intensity.
Here's the interesting part: in order to make white light you must mix, the red, green and blue colors. But it takes about 61% green, 31% red, and only 8% blue to make white light. This was determined by actual lab measurements.
This partially explains why white histograms don't show accurate exposure for each color. A white histogram basically measures green, since it is 61%. Blue only accounts for 8%, and most subjects don't contain much blue anyway. So the red channel is the most likely one to be blown out, since it contributes only 31%.
Rico,
The flare model is very intriguing.
I think the method that I proposed is relatively free from flare. It is mostly determined by the noise floor. Interestingly, no data so far has broken the 10 stop barrier. I think we have a couple of stops to go before hitting the limit suggested by the flare model.
slin100 wrote:
I've seen evidence somewhere, that the native color response of our sensors is more well-suited to a magenta-tinted light source. I've read of one photographer who has experimented heavily with his 10D and 20D and claims that he can achieve higher dynamic range with such a light source than with normal daylight.
The reasoning is that a tinted light source allows the three channels to independently achieve their maximum dynamic range.
I'm don't buy that at all, unless you're talking about just a slight tint, like what you'd get from a +/-10-point tint adjustment in ACR, to get the color balance perfectly dialed in to the sensor response. The fact that there are fewer red and blue pixels in a Bayer sensor has zero relevance to the color temp at which the pixels with each filter color will all output equal RAW values. In other words, the quantity of pixels has no connection to their individual output values (and the ratio between them) given a specific color of light. A given pixel's output value will not be affected by the color filter over any other pixel.
