F550 in good light.

Started Apr 18, 2011 | Discussions thread
Billx08 Forum Pro • Posts: 11,373
F550 in another light (part 1)

Wellington100 wrote:
. . .

I don't mind the F550's 16mp L size. Long before it or its images were available I wrote that because of the EXR sensor's design, 16mp was only useful as a marketing gimmick, that the really useful resolution would be 8mp, for three reasons. First, it should produce better images at 8mp than at 16mp because this was really its 'native' resolution, unlike 16mp Bayer sensors whose native resolution is 16mp, and they would produce inferior images when binned to 8mp.

Can you reconsider this assumption for our better clarification? My gut feeling here is that Bayer or not, there are 16m pixels, 16m Airy disks and 16m captures which are then binned into 8m pixels by some very low level firmware in the sensor. Its not a physical capture of 8m points of light, involving 8m Airy disks.

Ok, but to make what I want to describe easier to visualize (I hope) I'll resort to an example that some may think too odd to be useful. Suppose that a monochrome (to keep it simple) sensor's photosites are widely separated, as are the individual radiotelescopes in large radiotelescope arrays. They produce extremely high resolution images because the individual telescopes are widely separated (sometimes by thousands of miles), and if they were all pushed together to fit in a square mile, they'd all be essentially receiving the same signal, and the quality of the images would plummet. It's the same with camera sensors. You want to keep the photosites as far from each other as possible.

So suppose you fabricate a strange sensor, where the substrate isn't a silicon wafer but a football field instead. And suppose that there were 400 uniformly spaced football size photosites spread out on the field. That's analogous to a 400 pixel sensor. That's one sensor. Now consider another sensor that has only 200 football photosites uniformly spread out on the field. Each football is twice the size of the ones used in the 400 football sensor, so the resolution is lower because there are fewer of them, but because the size is greater, each football photosite collects twice as many photos per unit time as the smaller football photosites. So they're more sensitive to light, but ultimately don't capture any more photons in total than the 400 smaller photosites. Now consider the third and last sensor which uses 400 football photosites, but distributes them in adjacent pairs, so that from a distance this sensor is indistinguishable from the 200 photosite sensor. Each pair of the smaller photosites, because they're so close together, sees practically the same thing that the single larger photosite sees, so this 400 pixel sensor should function just as the real 200 pixel sensor does. Looking at this another way, suppose you had a team of super tiny nanobots running around atop a real CMOS sensor, and they had their own light sabers that could slice each photosite in half, and attach more wires so that each half's contribution could be combined and produce a single signal. That's somewhat similar to the EXR sensor's pixel pairs. Two adjacent photosites that under firmware control can be used collectively or individually.

What this means is that for all practical purposes, the EXR sensors in half resolution (M size) mode are little different than true 8mp sensors because due to the unique EXR layout, the photosite pairs used to make up a single photosite are adjacent to each other. In Bayer sensors, demosaicing the photosites to produce a single binned pixel uses parts of many photosites that are spread so far apart that they don't all see precisely the same part of the image, so the part of the image that they see is more blurred that what the more tightly joined EXR pixel pair sees. So assuming pixels large enough to avoid diffraction, a 16mp Bayer sensor will always produce more detail than a 16mp EXR image, because the EXR sensor really only has 8 million sensor clumps, and each clump (two photosites) are too close together to resolve nearly as much as if they were spread farther apart. So EXR sensors produce great M size images (5,6 or 8mp) but relatively poor L size images (10, 12 or 16mp), and Bayer sensors produce better 16mp images, but when binned to 8mp, they're not a match for the Bayer's 8mp images.


Post (hide subjects) Posted by
(unknown member)
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow