Nickname, 09 Feb 2020You both are a bit wrong. All three, simple bayer, quad, and nona; use 2x2 CFA. That grid make... moreThat would be what I was hoping for, but we'll see. For now we only have a trademark.
AnonD-754814, 09 Feb 2020Agreed, the debate is going nowhere. Let's wait and see What Samsung has to offer. Quad bay... moreYou both are a bit wrong. All three, simple bayer, quad, and nona; use 2x2 CFA. That grid makes what you can think of as a physical pixel of sensor. Quad bayer then use information from 2x2 grid of physical pixels to produce one pixel of image file, but can fall back on using 1:1 grid. Nona is able to use 3x3 grid of physical pixels to produce one pixel of image file, but can fall back to using quad or 1:1. In essence, nonacell is improvement of demosaicing algorithm, not of sensor, but as always marketing.
S Yu, 08 Feb 2020When I said 2x2 and 3x3 I meant how many photosites share the same color under the CFA, so con... moreAgreed, the debate is going nowhere.
Let's wait and see What Samsung has to offer.
Quad bayer is a improvement over normal bayer and Fuji X trans (6 to 1 CFA) is better than a Quad bayer so, in that logic 9*9 nonacell should at least be better than Quad bayer.
AnonD-754814, 07 Feb 20202*2 ? Even the quad bayer has 4*4(4 red 4 blue 8 green).
The bayering is 2 by 2 . But the co... moreWhen I said 2x2 and 3x3 I meant how many photosites share the same color under the CFA, so converted to bayer that means 4x4 and 6x6, though I fail to see the point in this conversion as we all know they're both ultimately RGGB bayer, the only difference is the number of photosites grouped together which is what I was stressing.
It's irrelevant whether "we are not wiser than them" or not, we are talking about technical details while they're doing business, engineers all love unlimited budgets but that's almost never the case, moreover Samsung could end up intentionally holding back more complex implementations for next year if they see the simplest binning as a big enough improvement for a generation, since the sensor itself is a huge jump.
100x zoom camera on a smartphone will be real innovation! I can't wait to see what's next!
Anonymous, 06 Feb 2020Why? Most photos are taken pretty quickly. And if the hardware is supposedly that crappy and o... more"Shot-to-shot time of several seconds is painful" - GSMArena, Lumia 1020 review
"8MP PureView photos are snapped in under a second, while 38MP shots might take 2-3 seconds" - GSMArena, 808 Pureview review
Didn't see one for the N8, but being 2 years older still, I expect it was even slower.
S Yu, 07 Feb 2020It's not 9x9 , I'm still assuming 2x2, but they'll need to jump through a few hoops after read... moreTo do 9*9 is to achieve stable high ISO to do better at low light and less color noise(better color and better sharpness).
S Yu, 07 Feb 2020It's not 9x9 , I'm still assuming 2x2, but they'll need to jump through a few hoops after read... more2*2 ? Even the quad bayer has 4*4(4 red 4 blue 8 green).
The bayering is 2 by 2 . But the color filter array is 4*4 . What I think is the number of color filter array really doesn't matter as long as it's well balanced with the bayering method. For 9 to 1 bayer the CFA can be 3*3(Which is unlikely), can be 6*6, 9*9, 12*12, 18*18 or anything which matches the best with nona bayer.
For example the quad bayer uses 4*4 CFA but pixel binning happens at 2*2 level.
So, it will be too early to say anything like it won't do any good.
I really don't have engineering level of knowledge on this. But like I said it's not impossible and clearly we are not wiser than them.
You should read about the Fuji X Trans sensor, see how it does the CFA distribution..
AnonD-754814, 07 Feb 2020There have been a dozen of bayer sensor with different color filter array with almost 50 diffe... moreIt's not 9x9 , I'm still assuming 2x2, but they'll need to jump through a few hoops after readout, during demosiacing, to make use of the additional color data. The sensitivity of 2x2 is the same as 3x3 as long as it's still a 9-1 bin, the only difference is potentially better color resolution. And still, assuming it's a two-tier HDR, not three-tier as it seems likely to introduce additional artifacts with limited benefits(two-tier could add 2-3ev DR, three-tier theoretically could add more but you can only prolong the shutter speed by so much before motion blur gets out of hand), using a 3x3 CFA sacrifices too much color data for...the main reason I could think of would be simplicity of software implementation, that's it, a technical compromise, not for squeezing more performance out of the hardware.
S Yu, 07 Feb 2020Ah ha, that I answered, as in I don't know the details, but IMO it amounts to a more effective... moreThere have been a dozen of bayer sensor with different color filter array with almost 50 different combinations altogether. Fuji already has 6*6. So, I don't think Samsung will make a mess with 9*9.
Anyway why would a quad byer be 3*3 ?
AnonD-754814, 07 Feb 2020There is no information that Samsung won't use 9*9 CFA. So, what are you shouting about ? Ad... moreThere was simply a misunderstanding. I already replied to him but the very long comment got blocked, I haven't written anything sensitive but I'm not entirely sure that the comment will get released, as with any moderating system.
viveksubhash, 07 Feb 2020I really don't understand why you would want 100x zoom.. You can't take a picture without a tripodYou can, actually, precisely because of the severe digital crop, it has far enhanced digital stabilization, assuming the 10x of the 100x is digital, then you have 99% of the frame(in area) that could be used for digital stabilization, which is essentially "floating" the crop area around the frame in the opposite direction of the movement of your camera. When your crop area could float the width/height of ~4.5 frames in any direction, that could make for *very* stable viewfinding, know that regular digital stabilization often work with a 1.2-1.3x crop, so that's just around 15% in screen width/height for the frame to float. That's generally enough for wide angle though.
Nick Tagataka, 07 Feb 2020Sorry if I confused you, I acknowledge that S20 will come with 9-1 Bayer sensor which Samsung ... moreAh ha, that I answered, as in I don't know the details, but IMO it amounts to a more effective subpixel sampling, the subpixel here meaning individual photosites, 9 of them forming a full pixel.
Just as a crude example, a group of 3x3 photosites in quad bayer means 4 red, 4 green, 1 blue; 2 red, 5 green, 2 blue; or 1 red, 4 green, 4 blue, so directly mapping the 9 to one pixel would be very stupid with a severe imbalance of color data. However, if it's allowed to look around and reference the data, weighted, of course, from neighboring photosites, then in the total 5x5 area it would have access to 9 red, 12 green, 4 blue; 6 red 13 green, 6 blue; or 4 red, 12 green, 9 blue, which would be far more balanced.
If even I could think of a sampling method like this then surely there could be a more complex and sophisticated demosiacing algorithm.
The single shot HDR could directly be reused from regular quad bayer, whichever way you group the 9 photosites, it will have 4/5 fast/slow photosites, which is very close to the 2/2 in quad bayer and surely is not a major issue with a little tweaking. Instead, using 3 shutter speeds for 9 pixels in a 3/3/3 configuration would seem more complex.
The Last Oracle, 07 Feb 2020The fact is, that, the Bayer Pattern is the single largest hindrance to image quality, includi... moreYou have zero clue on how modern day Bayer sensors coupled with 2020 Algorithms and image processing works.
The Last Oracle, 07 Feb 2020The fact is, that, the Bayer Pattern is the single largest hindrance to image quality, includi... moreThe Bayer pattern isn't intended for anything you discuss. It's there to tackle one of the biggest issues with mobile cameras, accuracy in the tiny lenses and quality consistency in the tiny sensor. That problem is magnified by the varying behaviour of coloured light. Getting 9 unique sample points for each pixel provides raw data that are indefinitely more useful than any guesswork "neural learning" and software can ever achieve.
Sounds like a Chinese copy of a patented thing from a big brand :-)
Adidas —> Adiads
Nanocell —> Nonacell
The fact is, that, the Bayer Pattern is the single largest hindrance to image quality, including resolution, low light, sharpness, and responsible for introducing all sorts of false patterns that greatly affect image quality. Also over-degreed fools quoting Nyquist limits (which frankly, should be limited to sampling audio), and other theories, which have scant application to lens and sensor optics, make understanding of final image quality highly blurred. This is why the l16 camera was so brilliant in theory and rather mediocre in practice. I guess a Pixel 3 or 4 image, upressed (by way of any good upressing software), would probably produce more impressive images than the l16. The same applies to the Nokia 9 too. Stating that its FIVE x12 MP cameras, collect up to 10 times more light than a single smartphone camera color sensor, was great in theory, but not really true in practice (and that's why regular smartphone cameras with Night Mode or Low Light Mode captured much more impressive night photos). A single sensor with a hybrid-CFA can probably produce 2-5 times as much detail and sharpness as presently possible, and OIS and multi-frame can easily compensate for about 2-3 times more, creating an image that would have, without OIS and HDR, only been 2/th to 1-15th as detailed and as impressive in low light and all other situations.