I think the key to understanding the way they construct the simple Low-Light ISO score is this paragraph:
Quote: An SNR value of 30dB reflects an excellent image quality. Low-Light ISO is then the highest ISO setting for the camera such that the SNR reaches this 30dB value while keeping a good dynamic range of 9EVs and a color depth of 18bits.
So (and I may be wrong...) the raw number means, for example in the RAW output from the camera:
- for the D700, at an ISO setting of 2303, the signal-to-noise ratio still reaches 30db, whilst still holding onto a dynamic range of 9EV and a color depth of 18bits;
- for the K20D, only at a lower ISO setting of 639 can the same parameters for SNR, dynamic range and color depth apply.
- ditto for the D90 at 977, the EOS450D at 692 etc.
It seems quite a tough test, actually, because those ISO's are quite low. It is even saying that the D90 can only deliver those results at a comparatively low ISO.
For the K20D the real comparisons should be within the APS-C space, and here the results for the K20D aren't much different from any of the newer Canon's. Only the D90/D5000 break away from the rest of the APS-C pack, so far.
APS-C size sensors are fundamentally quite boxed in when it comes to achieving good SNR whilst holding onto dynamic range and color depth, I think. Certainly when compared to full-frame.
And DXO do make it clear that their benchmark has nothing to do with all of the matters that may impact optical or camera quality:
General questions
I'm not too sure how the concept of 'detail retention' works its way into their calculations though. One would need to define what detail retention means, in terms of what a camera sensor does. I guess if anything, detail retention must also be a function of SNR, dynamic range and color depth?
Anyway, don't shoot the messenger. I think there stuff is useful because it's a methodically assembled dataset. But I'm no expert on sensors, SNR etc etc.
Hopefully this sort of stuff will encourage Pentax/Samsung to redouble their efforts to tweak their k7 firmware etc to do better against the Canikons.
Maybe even push them a bit harder along the way to cooking up a full frame sensor too, to match Canikon/Sony.