Understanding RawDigger's New Over- and Under-Exposure Facility

Here is my understanding of the behavior of the Over- and Under-exposure indicators now present in RawDigger (version 0.9.13 beta). This information is derived from my own experimentation with the program and, more importantly, from some very excellent correspondence with Alex Tutubalin of RawDigger, for which I am very grateful. Needless to say, I am responsible for any errors. Alex has kindly offered to try to monitor this thread to answer any questions that may arise.

Auto Overexposure Point:

The Auto Overexposure Point is determined from the histograms as follows: Underexposed images are assumed to have histograms that tail off on the high end, so overexposure is assumed to exist if the histogram peaks on and at its high end. Such a peak is defined if the highest ADU level, L, with a non-zero pixel count has a count that exceeds that of the preceding level, L-1, by roughly 100 pixels. The Overexposure Point then becomes L-s, where s is the Sensitivity.

The Sensitivity, s, is L*p, where p is the percent chosen in the preferences. Thus, the Overexposure Point is L-L*p = L(1-p). If, then, p is set to 1%, the overexposure point is L*0.99.

Because some Nikon cameras have saturation points that differ spatially across the sensor, if the above condition is not met, the comparison is made over N levels, where N is determined as s=L*p. [See note 1]

If neither of the above conditions is met (i.e., there is no overexposure), the Overexposure Point (Saturation Point) is set at T-b, where T is the "theoretical maximum" and b is the Black Level [See note 2]. This saturation setting appropriately prevents any overexposure blinkies from being displayed while allowing the "Offset from Saturation" to work correctly for determining underexposure.

Manual Overexposure Point:

You can see the behavior of the Overexposure Point setting by using the Manual Level (for all channels) and keeping the "Reset Manual levels on file load" checked. If you analyze a clearly overexposed file, you will see this level is set to the highest ADU shown on the histogram (L) less s=L*p, where p is your selected percentage. And if you examine an underexposed file, you will see this manual level is set to your highest possible ADU figure (T-b). [See note 3]

Be aware that there is a slight anomaly in the behavior of the preference dialog in making these settings. When you select a percentage, the Auto button is automatically activated even if you had selected Manual setting. After setting the percentage, then, you must (re)select the Manual setting. The percentage (p) you've selected, however, will apply to the calculation of the manual Overexposure Point.

If you've checked "Reset Manual level on file load," it doesn't make any difference what figure you enter in the Manual level box; this figure will be recalculated automatically as described above when you load your image. If the "Reset" box is not checked, however, the figure you enter becomes a hardwired Overexposure Point.

Underexposure Point:

The "Offset from Saturation" setting does just what it says: it sets the Underexposure Point at the selected number of EV below the Saturation Point determined above. Alex Tutubalin suggests that he uses Auto Overexposure with an EV-Offset equal to his camera's "working" dynamic range. I've tried this on a good number of images (using 1%), and it does indeed seem to work quite well.

A Manually entered Underexposure Level is, as is to be expected, relative to (over and above) the black point.

Notes:

Note 1: I am confused by this condition since it appears to be determining L through the use of L. Perhaps there is some sequential process involved that resolves this conundrum. Some clarification is needed, at least for some Nikon users.

Note 2: The Theoretical Maximum is typically 2^(ADC bit count)-1 for raw files without a tone curve and that value transformed by the tone curve for cameras with a tone curve. However, some configurations for some cameras can have a theoretical maximum lower than this value. The 12-bit E-M5, for example, has 2^(ADC bit count)-1 = 4095. But the value of T is 4065 for ISO 200 and then jumps to 4093-4095 for higher ISOs. The Auto black point for the E-M5 is b = 254 to 255. E-M5 values for T-b would, therefore, be around 3811-3815 at ISO 200 and 3839-3841 for higher ISOs.

Note 3: For all the images I've examined showing overexposure, L = T-b, and I think this will typically be the case for cameras that have sensors with spatially constant saturation levels. It may well be that there are other configurations with which I am not familiar where this might not be the case. Again, further information is desirable.

--
gollywop