50 Mpixels? Hah! The new bar is 56 Mpixels

With megapixels going into the stratosphere, I'm wondering when it will be possible to put a 170-degree fish eye lense on a camera and then digitally zoom to the user's desired frame of reference... Of course, zoom lenses will still have their place but this sort of technology would re-define "everyman" camera.

I think we''re in violent agreement, Steve. I was saying that indeed the AD is 14b in the XSi. There is 14b of data but there is not 14b of *information.* This is for a properly exposed image at low ISO. Up the ISO and the noise moves left. My contention is that bit 0 and maybe bit 1 of each sample do not just contain some noise they contain *only* noise. (By the way, noise isn''t so bad. Noise in the LSBs provides some dither, which makes certain smooth color graduations look better than if it were absent.) By the way, I googled on using imaging sensors as noise sources, and once again, I have been thwarted in my efforts to have an original thought. There is an open source project called "lavarnd" that uses cheap webcams for precisely this purpose. However, in this case, they put the sensor in the dark, insuring that all readings come purely from thermal noise.

Dave J: I know for a fact that there's a 14-bit A/D in the Canon XSi and I'm not at all surprised if there's some noise in the LSBs of the RAW files. I'm anxious to see what the new Digic 4 processor in the Canon 50D does with these 14-bit values. Some sort of multishot averaging mode would be interesting. I doubt that the Leaf AFi-II has a 16-bit A/D in it. Too much money and too much color depth for a 12-to-13-stop dynamic range.

Thanks, Steve. Sorry if I missed your intent. I can be a dork for technical pedantry and I''ve gone nuts for digital photography lately. There''s an interesting site here: clarkvision.com/imagedetail/eye-resolution.html that has an nice discussion of the capabilities of the human eye. An interesting complication is the fact that vision is a composite process of the brain and eye, and the brain is constantly stitching together data from the eye to make (the illusion of?) more complete images. As for the bitness, I''d be surprised if they just pad with zeroes. I think what these cameras have are D-A converters that do indeed provide 16b of output, but the sensor does not really provide 16b resolution worth of input -- meaning their are numbers there, but they don''t mean much. I have verified the same phenomenon in the consumer cameras. For example the Canon XSi has "14b raw" and also shows about 12-13 stops of dynamic range. Real raw files show that there are indeed numbers in those LSBs -- it''s just that they don''t actually convey much info. It''s marketing. In fact, I was pondering the other day of using my camera at high ISO aimed at a uniform gray object as a potential source of high-quality true random numbers. Why use one noise diode when you can harness the work of 10.1 million?

Dave J, I'm pretty sure that if you reread my writing, I say that consumer-grade cameras aren't likely to need 56 Mpixels. A 56-Mpixel camers is an achievement nevertheless. My reading indicates that human visual acuity is equivalent to somewhere around the 16-20 Mpixel range (we don't really see pixels, so the conversion must be approximate), which is currently the top end for pro and prosumer dSLRs and roughly 2x beyond the reach of today's best point-and-shoots. As for the 16-bit TIFFS, that's the file format you must use for 12-bit/pixel data. I suspect the four LSBs are zeroed in this camera. You can't blame the camera for using a standard file container. The point of that line in my original post was to point out the immense file sizes required. Happy days for flash vendors. Everything you argue is correct, but I didn't intend the meanings you construed. However, thanks for reading my stuff so closely!

Steve, I'm surprised you're so excited by pixel counts. Personally, I could do with quite a few less -- especially if I get better noise performance at high ISO in return. This camera, intended for the studio, handles only ISO 50-800, and I have a feeling its designers did not spend much effort on the high end of that ISO range. A few other drawbacks: 1 fps, 4.8 lbs, positively enormous files. (Though their lossless-compressed RAW format is "only" 78MB -- surprisingly more efficient than the 16b TIFF you quoted.) Finally, they're giving you 16bits of data, but only 12 stops of dynamic range, which tells me that four of those bits probably are not much better than pure noise.