I have long been looking for how to present the results of the photon transfer function analysis that I perform on sensors. Below is my now-standard presentation, for the Hasselblad X2DII.
This is a normalized presentation, with the vertical axis normalized to sensor height in pixels using the same normalization factor (1600/height, or height/1600, depending on how you look at it) that Bill Claff uses on his Photons to Photos website. The black line marks Bill’s photographic dynamic range threshold. You can read PDR directly off the chart by noting the x-axis locations where the curves cross the black line. You can see how the ISO 200 curve, which is the lowest ISO with high conversion gain in this camera, tracks with everything else at the right side of the graph, where photons are plentiful, has the same Claff PDR as ISO 100, and does better than ISO 100 for areas darker than Bill’s threshold. There’s some funny business going on with the ISO 12400 curve; it looks normal in bright regions, but is better than you’d expect in dark regions. This is indicative of digital processing in the camera to lower noise.
Here’s another way to look at the same data. I’ve plotted photographic dynamic range for a set of different SNR thresholds.
The purple line is for Bill’s PDR threshold. The lines above it are for darker thresholds, and the lines below it are for lighter thresholds. You can see that the dual gain structure of the sensor has little effect with log2(SNR) = 5, 6, or 7. Those numbers correspond to signal to noise ratios of 32, 64, and 128. In those regions photon noise is the long pole in. the tent, and the read noise improvements provided by dual conversion gain don’t help. But at thresholds darker than Bill’s log2(SNR) = 3.3, the SNR at ISO 200 is better than that at ISO 100, which in the days before the Aptina dual conversion gain trick would have seemed passing strange.
One thing that wasn’t obvious on the upper graph that is easily seen on the lower one is that there is some processing going on at ISO 12400 that is improving the shadow SNR at levels below the Claff threshold.
What you you all think? Is this a useful way of presenting this data?
CarVac says
I prefer the information density of the top plot; the bottom one loses information by discretizing and limiting the range to that of the chosen SNR thresholds.
Neither is *that* useful for comparing cameras in one plot, either.
I do wish PhotonsToPhotos would make the top plots available (I know DxO does, but they don’t keep the scale consistent).