Godfrey DiGiorgi wrote: > On Jul 5, 2006, at 12:11 PM, Toralf Lund wrote: > > >> So, essentially what you are saying is that you want to retain the >> data >> "exposed to the right" past the gain phase because that can only scale >> the values in a linear fashion, while the RAW conversion is non- >> linear? >> (And non-linear in such a fashion that an input where the low-range of >> values are "stretched" is advantageous.) >> ... >> One minor point, though: Do you actually know that the CCD has 4096 >> different values? I mean, the A/D gives you that, and you obviously >> expect that to be matched to the CCD, but I've been thinking that >> it may >> make sense (partly due to the gain etc. involved) to keep the sensor >> resolution slightly higher that the one of the A/D. If it is set up >> like >> that, obviously you want to do as much of the scaling/"exposure >> correction" as you can in the analogue domain, as you'll retain >> more of >> the dynamics in the digital output that way... >> > > > The data output by the Pentax DSLR sensor into the RAW format file > contains photosite values in the range from 0-4095, that is, in > numbers bounded from 0 to 2e12. That data is the output > characteristic of a linear sensor device and requires gamma > correction to be transformed into a tonal scale normal for the human > eye to appropriate. > The rest of what you write is incomprehensible and/or irrelevant. > > I was talking about the number of different voltage levels that may be output from the sensor itself. That number is not necessarily 4096; it is the analogue-to-digital converter that has 4096 different values. This distinction is far from irrelevant. Unless the actual sensor can produce more than 4096 different levels, your whole exercise is quite pointless at ISO settings much higher than the native rating of the sensor. Heck, at low-light conditions, there is not even point in metering at all, if you are going gamma correct and/or "scale" the data as part of our raw conversion anyway.
I think you'll understand why if you consider the following "experiment" for a while: Get a digital camera with a native ISO of 100. Turn the ISO dial to 100, and set shutter/aperture so that you get correct exposure, and start firing away. Check the maximum pixel value. Reduce the amount of light by 50%. Don't update the camera settings. Try again. What is the pixel value range like now? Now switch to ISO 200, and repeat. The big question is whether the last step actually changed the number of different pixel values, or just replaced the range 0,1,2,3...2047 with 0,2,4,6...4094. Like I said, think about it for a while... - Toralf -- PDML Pentax-Discuss Mail List PDML@pdml.net http://pdml.net/mailman/listinfo/pdml_pdml.net
