Prev: SI Facescape
Next: FF camera with mirrorless design
From: David J Taylor on 30 Apr 2010 10:35 > But the question is, what the performance of the sensor is. It may well > be the case that the lens sucks and the AA filter additionally > contributes to blur everything, but this is a separate issue. > > The key question is, can the sensor resolve pixel-level changes of > colour. A Bayer sensor *cannot*, while a Foveon sensor or other full- > colour imaging device which does not have to interpolate the data from > adjacent pixels can. > > And by the way, with Bayer cameras you are forced to design the AA > filter for the lower colour resolution of the sensor (1/2 of the sensor > resolution) if you want to avoid colour aliasing. With a full-colour > sensor you can instead design the AA filter for the full resolution of > the sensor. > -- > > Alfred Molon The system performance is what matters in the end, everything from subject, atmosphere, optics, AA & IR filters, sensor, processing, display, viewing environment (e.g. light-level, vibration), eye optics, image on retina, brain performance. Probably even more factors. Of course the sensor matters, but it needs to be considered along with the rest of the system. As the eye can't resolve pixel-level changes of colour, for use with human perception to make a sensor which can do so could be considered over-engineering. For typical use, the Bayer sensor has adequate performance at an affordable price, and is certainly not a linear factor of two down in sensor resolution. Otherwise you would need a 17MP Bayer sensor to equal a 4.3MP Foveon one, and that doesn't match up with most reports. Of course, you can probably find special-case exceptions, and if your imaging needs fall within those exceptions, you choose the most appropriate tool for the task. Cheers, David
From: Alfred Molon on 30 Apr 2010 15:15 In article <hreprv$vl6$1(a)news.eternal-september.org>, David J Taylor says... > The system performance is what matters in the end, everything from > subject, atmosphere, optics, AA & IR filters, sensor, processing, display, > viewing environment (e.g. light-level, vibration), eye optics, image on > retina, brain performance. Probably even more factors. Of course the > sensor matters, but it needs to be considered along with the rest of the > system. Tests show that resolutionwise the Sigma 4.6MP camera is on par with 7- 8MP Bayer cameras. > As the eye can't resolve pixel-level changes of colour, for use with human > perception to make a sensor which can do so could be considered > over-engineering. That obviously depends on the viewing distance, enlargement etc. Or do you think that 60MP medium format cameras are overengineered because the human eye cannot see so much fine detail? > For typical use, the Bayer sensor has adequate performance at an > affordable price, and is certainly not a linear factor of two down in > sensor resolution. I never claimed that. -- Alfred Molon ------------------------------ Olympus E-series DSLRs and micro 4/3 forum at http://tech.groups.yahoo.com/group/MyOlympus/ http://myolympus.org/ photo sharing site
From: Alfred Molon on 30 Apr 2010 15:38 In article <gXACn.133291$mn6.68015(a)newsfe07.iad>, Martin Brown says... > It doesn't only capture 1/3 of the information. It is actually quite > tricky to put an exact number on the amount of information that it does > capture. But based on the 0.30, 0.59, 0.11 weights for R,G,B a Bayer > quad captures (0.3+2*0.59+0.11)/4 = 0.40 or 40% of the information. > > Experience has shown that as long as the luminance signal has close to > full resolution (and Bayer demosaicing ensures that) the lower > resolution chroma information is invisible to the human eye except in a > few special test cases. TV broadcast has exploited this for decades. While the green channel is a good approximation of luminance, the red and blue ones are not. Regrettably there are only 50% of green pixels in a Bayer array and that is not close to 100%. > Actually it is after a fashion almost optimal. It measures the essential > components needed to reconstruct a full colour representation of the > image in an efficient manner. In half of the pixels Bayer does not even capture a good approximation of luminance. And it is impossible to reconstruct the colour information at the pixel level without some interpolation which introduces errors. > Not when the sensors are so much smaller and seem to have nasty > artefacts if there are specular highlights in the FOV. Still Foveon has approx. 50% more effective resolution than Bayer sensor with the same pixel count. > >> interpolation is *not* guessing. > > > > It is, because the value in the middle does not depend on the > > neighbouring values. > > It does depend very strongly on the adjacent pixels that is true unless > the image is undersampled which is what the antialias filter prevents. That depends on the image. If the image has large blobs of homogeneous colour fine, if not Bayer perfoms very poorly. In any case, what is the point of continuing this discussion. I keep repeating the same things all the time. Look at the reviews: Foveon 4.6MP performing as a Bayer 8MP sensor. Do we need an endless thread? If you don't like Foveon cameras, don't buy them. -- Alfred Molon ------------------------------ Olympus E-series DSLRs and micro 4/3 forum at http://tech.groups.yahoo.com/group/MyOlympus/ http://myolympus.org/ photo sharing site
From: David J Taylor on 30 Apr 2010 17:20 "Alfred Molon" <alfred_molon(a)yahoo.com> wrote in message news:MPG.26454ac8abfabdf598c2c5(a)news.supernews.com... > In article <hreprv$vl6$1(a)news.eternal-september.org>, David J Taylor > says... > >> The system performance is what matters in the end, everything from >> subject, atmosphere, optics, AA & IR filters, sensor, processing, >> display, >> viewing environment (e.g. light-level, vibration), eye optics, image on >> retina, brain performance. Probably even more factors. Of course the >> sensor matters, but it needs to be considered along with the rest of >> the >> system. > > Tests show that resolutionwise the Sigma 4.6MP camera is on par with 7- > 8MP Bayer cameras. > >> As the eye can't resolve pixel-level changes of colour, for use with >> human >> perception to make a sensor which can do so could be considered >> over-engineering. > > That obviously depends on the viewing distance, enlargement etc. Or do > you think that 60MP medium format cameras are overengineered because the > human eye cannot see so much fine detail? > >> For typical use, the Bayer sensor has adequate performance at an >> affordable price, and is certainly not a linear factor of two down in >> sensor resolution. > > I never claimed that. > -- > > Alfred Molon Thanks. Alfred. Given the choice of today's 4.6MP Foveon camera or a 10-12-15MP Bayer camera, there appears to be an obvious resolution gain in the Bayer camera, plus the better high-ISO performance others have reported. 60MP would almost certainly be overkill for what I do, and for what many people do with their photography. You choose the tools according to your needs. Thanks for your clarification in your last paragraph. It appears that the resolution loss with Bayer is around 1.3 (linear) from your first paragraph. Cheers, David
From: nospam on 30 Apr 2010 17:46
In article <MPG.2645503552d7b85d98c2c7(a)news.supernews.com>, Alfred Molon <alfred_molon(a)yahoo.com> wrote: > In half of the pixels Bayer does not even capture a good approximation > of luminance. And it is impossible to reconstruct the colour information > at the pixel level without some interpolation which introduces errors. very small errors which most of the time will never be noticed. > > > Not when the sensors are so much smaller and seem to have nasty > > artefacts if there are specular highlights in the FOV. > > Still Foveon has approx. 50% more effective resolution than Bayer sensor > with the same pixel count. false. > > >> interpolation is *not* guessing. > > > > > > It is, because the value in the middle does not depend on the > > > neighbouring values. > > > > It does depend very strongly on the adjacent pixels that is true unless > > the image is undersampled which is what the antialias filter prevents. > > That depends on the image. If the image has large blobs of homogeneous > colour fine, if not Bayer perfoms very poorly. bayer does poorly on red/blue test charts. otherwise, it does very well. > In any case, what is the point of continuing this discussion. I keep > repeating the same things all the time. Look at the reviews: Foveon > 4.6MP performing as a Bayer 8MP sensor. bayer was 10 mp about 4 years ago. canon's entry level camera *now* is 18mp. nikon's high end is 24mp. even if you think 4.6 looks like 8, it's way behind. > Do we need an endless thread? If > you don't like Foveon cameras, don't buy them. most people don't. |