Don writes: >But, sorry. In the interest of disclosure, I've never actually instrumented all this. So the whole argument is totally subject to refutation by someone who has actually done the research, say, by evaluating CCDs in a cryostat in an RF-shielded room, then repeating in ambient over a statistically significant period / number of cycles. >My gut says I'm willing to bet anyone a quarter that most of the production devices are pretty raunchy at least in the low order 2 bits or so. My experience at testing to any degree has been limited to shooting a Gretag scale, checked against a densitometer, and the examining the densities coming out of the camera and the response curve was certainly not linear (on the cameras I used).. nor was it consistent! It seems the clever beetles in the camera do some intelligent interpretation and slap the data through a few algorithms before laying down the values. This is the bit that offends my scientific sensibilities, the fact that I cannot get a consistent result therefore cannot know with any surety how the sensor will respond >That says nothing about the inherent noise floor of inexpensive CCDs, which *will* affect the linearity and monotonicity of the output that's another thing - the response and noise varies with the heat. those little semiconductors are pretty wavy in their responses when heat and light are both present, which invariably they are, and while onboard noise cancelling based on temperature monitoring *may* work to a certain extent, it's going to act a bit like a bandpass filter and change the values of the light recorded. >All that's going on is that the sampling times are varied and noise cancellation algorithms are applied to the raw data. But would one actually imagine that one's image encoding is "linear", just because someone did a statistical noise cancellation on the data? exactly. that's an after the fact modification to render it close(r) to linear. Anyone who has had to match a pair of semiconductors for close analogue response will know, going through a box of 100 transistors/diodes/scr's or whatever, matching just a pair can be a right PITA, matching a couple of million sounds like a nightmare! Of course 'mapping' their responses and applying a filter after the fact might be easier, but again, temperature/time/etc variations are going to make their analogue responses different when they fall outside the calibration/mapping, and there will be drift and failures with time. Having said all that you *can* get to know your camera - maybe not absolutely, but film of course was no better, just different. Each batch had to be tested and that was a job in its self.. however once a batches response was known with processing at whatever dev/temp/time, you pretty much knew what you were doing. k
