5 ms·

(As an aside, can you talk more about the Bayer detail-destroying filter? I love this stuff.)

by joshu 15y ago

aeturnum 15y ago
Most CCDs extract color information using a Bayer Filter[1] over them. Each square is a pixel in the megapixel count (a 1 megapixel sensor would have 250k red, 250k blue and 500k green photo sites). Algorithms are used to get RGB values for each pixel in that setup (luminance is extracted from the green sites). However, all the algorithms for the Bayer filter have a tendency to generate Moiré patterns[2]. To combat that, most sensors have anti-alias filters in front of them. This reduces the size of detail that can resolved by a sensor, but it also prevents some of the ugly consequences of the Bayer filter. Some companies have experimented with other sensors or other patterns. The new Fuji X-PRO1 is using a non-Bayer filter, which should be interesting if it works well (it also does not have an anti-alias filter). There are also Foveon sensors, which use the selective permeability of silicon to different kinds of light instead of a pattern of photo sites.[3] [1] http://en.wikipedia.org/wiki/Bayer_filter http://en.wikipedia.org/wiki/Bayer_filter [2] http://en.wikipedia.org/wiki/Moir%C3%A9_pattern http://en.wikipedia.org/wiki/Moir%C3%A9_pattern [3] http://en.wikipedia.org/wiki/Foveon_X3_sensor http://en.wikipedia.org/wiki/Foveon_X3_sensor
ajkessler 15y ago
He's talking about anti-aliasing filters (http://en.wikipedia.org/wiki/Anti-aliasing_filter http://en.wikipedia.org/wiki/Anti-aliasing_filter), which essentially blur the image enough to mask moire (http://en.wikipedia.org/wiki/Moire http://en.wikipedia.org/wiki/Moire). Because this is the internet, I feel obliged to point out the inaccuracies of some of his other statements: >>Film was and is to some degree still capable of more resolution, more dynamic range, better color fidelity and less prone to weird errors in the way that a Bayer sensor does<< First of all, film and digital sensors of comparable sizes have nowhere near the same amount of resolution, dynamic range, or color fidelity. Digital sensors simply murder film in every category. (You can make the argument that B&W film still has some an edge in terms of dynamic range, but considering you're matching it against RGB sensors, I'm not sure that's fair.) Even 8x10 film is now out-resolved by medium format backs. In terms of color fidelity, there is no film emulsion that will get you more accurate results than properly color-managed digital equipment. Which is exactly why museums and people who care about accurate color reproduction use digital capture. Finally, the anti-aliasing filters aren't exactly a big deal. Yes, they degrade image quality. Yet, even with them on most consumer sensors, those sensors still massively out-resolve, and produce sharper images, than comparable film. Also note that most medium format backs don't have AA filters. So, while somebody might like using film (or may need to in certain circumstances) the arguments presented aren't exactly the best justifications for doing so.
- gchpaco 15y ago
  Just 'cuz I can't let this pass without comment, same as you can: let's take a nice color film and compare this. Say, Ektar 100; in the past I might have used Ektar 25 for this but 100 is a nice speed good for outdoors shooting or controlled lighting conditions. I'm using a Kodak film here both because we're talking about Kodak and because they offer superb information on their products. I'm using a color film because technical B&W films exist and are superb for their application but are not really suitable for general photography, plus are more directly comparable to digital. Film is an analog medium, so we need some way of measuring resolution. The accepted way is line pairs per mm, or cycles per mm; a dark line next to a light line. A Bayer sensor because of the way the color filter works requires four linear pixels to resolve this same object. Now Kodak's papers on the subject indicate that Ektar's color response desynchronizes around 20 lp/mm and reach 20% contrast at around 65 lp/mm for the red channel and 80+ for the blue and green channels. Since we're dealing with black and white lines here, we use the blue and green filters as they will still be visually distinguishable. We can argue whether further than 20% contrast is relevant but it seems like a decent enough stopping place. Fine. So the film can resolve 80 lp/mm. Actual gate size varies a little but generally it is considered to be 36mm x 24mm. So that translates to 11520 x 7680 or an 84 megapixel camera, under perfect conditions. In realistic use because of diffraction you are unlikely to get 80 lp/mm on the film, the film is not likely to be flat, you probably can't hold the camera perfectly rigidly, etc. It is generally accepted by archivists that 4000 DPI scanning is good enough for archival, which translates to about a 5700 x 3800 image or about 20 MP. In 8x10 that translates to about 1.2 gigapixels, although when you get to that point 4K DPI is probably overkill due to the diffraction limit reached at typical 8x10 working apetures. As for dynamic range, Kodak doesn't show when Ektar 100 shoulders off but it has minimum 11 stops of dynamic range. I've used Ilford's XP2 when shooting a wedding before, and that has a bizarre shoulder with something like 13-15 stops of dynamic range. Digital cameras are typically more like 7 without using HDR, and HDR is not well suited to anything that isn't completely static. Finally we get to color fidelity. Kodak worked very, very hard to be as precise as possible with their films and largely succeeded; you can find in the Ektar 100 tech document precise filter recommendations to cancel out the oddity of florescents if you know what sort of florescent you're working with. If you are fortunate enough to use flash, all flashes are designed to emit a precise blackbox radiative spectrum that corresponds to daylight, where the films are balanced. EPP, which is a long obsolete reversal film that I have worked with, was used a lot in catalog photography where it was vitally important that the color used in the catalog must match precisely the color that was photographed, and it was very good for that. Digitals are theoretically more flexible and thus have the potential for better color fidelity, but to do that you basically need to know precisely which lights you are using and at that point you're in the same boat as film users. Now, is all this really compelling reasons to go to film? Not really. But I spend all day every day working in front of a computer and I'll be damned if I'm going to spend my hobby time calibrating my monitors and messing with scanners and printers (each in a dead heat for most miserable fucking peripherial ever), so I do traditional darkroom stuff, occasionally including some color work.

(As an aside, can you talk more about the Bayer detail-destroying filter? I love this stuff.)

by joshu 15y ago

(As an aside, can you talk more about the Bayer detail-destroying filter? I love this stuff.)

Most CCDs extract color information using a Bayer Filter[1] over them. Each square is a pixel in the megapixel count (a 1 megapixel sensor would have 250k red, 250k blue and 500k green photo sites). Algorithms are used to get RGB values for each pixel in that setup (luminance is extracted from the green sites). However, all the algorithms for the Bayer filter have a tendency to generate Moiré patterns[2]. To combat that, most sensors have anti-alias filters in front of them. This reduces the size of detail that can resolved by a sensor, but it also prevents some of the ugly consequences of the Bayer filter. Some companies have experimented with other sensors or other patterns. The new Fuji X-PRO1 is using a non-Bayer filter, which should be interesting if it works well (it also does not have an anti-alias filter). There are also Foveon sensors, which use the selective permeability of silicon to different kinds of light instead of a pattern of photo sites.[3] [1] http://en.wikipedia.org/wiki/Bayer_filter http://en.wikipedia.org/wiki/Bayer_filter [2] http://en.wikipedia.org/wiki/Moir%C3%A9_pattern http://en.wikipedia.org/wiki/Moir%C3%A9_pattern [3] http://en.wikipedia.org/wiki/Foveon_X3_sensor http://en.wikipedia.org/wiki/Foveon_X3_sensor

ajkessler 15y ago

He's talking about anti-aliasing filters (http://en.wikipedia.org/wiki/Anti-aliasing_filter http://en.wikipedia.org/wiki/Anti-aliasing_filter), which essentially blur the image enough to mask moire (http://en.wikipedia.org/wiki/Moire http://en.wikipedia.org/wiki/Moire). Because this is the internet, I feel obliged to point out the inaccuracies of some of his other statements: >>Film was and is to some degree still capable of more resolution, more dynamic range, better color fidelity and less prone to weird errors in the way that a Bayer sensor does<< First of all, film and digital sensors of comparable sizes have nowhere near the same amount of resolution, dynamic range, or color fidelity. Digital sensors simply murder film in every category. (You can make the argument that B&W film still has some an edge in terms of dynamic range, but considering you're matching it against RGB sensors, I'm not sure that's fair.) Even 8x10 film is now out-resolved by medium format backs. In terms of color fidelity, there is no film emulsion that will get you more accurate results than properly color-managed digital equipment. Which is exactly why museums and people who care about accurate color reproduction use digital capture. Finally, the anti-aliasing filters aren't exactly a big deal. Yes, they degrade image quality. Yet, even with them on most consumer sensors, those sensors still massively out-resolve, and produce sharper images, than comparable film. Also note that most medium format backs don't have AA filters. So, while somebody might like using film (or may need to in certain circumstances) the arguments presented aren't exactly the best justifications for doing so.

gchpaco 15y ago
Just 'cuz I can't let this pass without comment, same as you can: let's take a nice color film and compare this. Say, Ektar 100; in the past I might have used Ektar 25 for this but 100 is a nice speed good for outdoors shooting or controlled lighting conditions. I'm using a Kodak film here both because we're talking about Kodak and because they offer superb information on their products. I'm using a color film because technical B&W films exist and are superb for their application but are not really suitable for general photography, plus are more directly comparable to digital. Film is an analog medium, so we need some way of measuring resolution. The accepted way is line pairs per mm, or cycles per mm; a dark line next to a light line. A Bayer sensor because of the way the color filter works requires four linear pixels to resolve this same object. Now Kodak's papers on the subject indicate that Ektar's color response desynchronizes around 20 lp/mm and reach 20% contrast at around 65 lp/mm for the red channel and 80+ for the blue and green channels. Since we're dealing with black and white lines here, we use the blue and green filters as they will still be visually distinguishable. We can argue whether further than 20% contrast is relevant but it seems like a decent enough stopping place. Fine. So the film can resolve 80 lp/mm. Actual gate size varies a little but generally it is considered to be 36mm x 24mm. So that translates to 11520 x 7680 or an 84 megapixel camera, under perfect conditions. In realistic use because of diffraction you are unlikely to get 80 lp/mm on the film, the film is not likely to be flat, you probably can't hold the camera perfectly rigidly, etc. It is generally accepted by archivists that 4000 DPI scanning is good enough for archival, which translates to about a 5700 x 3800 image or about 20 MP. In 8x10 that translates to about 1.2 gigapixels, although when you get to that point 4K DPI is probably overkill due to the diffraction limit reached at typical 8x10 working apetures. As for dynamic range, Kodak doesn't show when Ektar 100 shoulders off but it has minimum 11 stops of dynamic range. I've used Ilford's XP2 when shooting a wedding before, and that has a bizarre shoulder with something like 13-15 stops of dynamic range. Digital cameras are typically more like 7 without using HDR, and HDR is not well suited to anything that isn't completely static. Finally we get to color fidelity. Kodak worked very, very hard to be as precise as possible with their films and largely succeeded; you can find in the Ektar 100 tech document precise filter recommendations to cancel out the oddity of florescents if you know what sort of florescent you're working with. If you are fortunate enough to use flash, all flashes are designed to emit a precise blackbox radiative spectrum that corresponds to daylight, where the films are balanced. EPP, which is a long obsolete reversal film that I have worked with, was used a lot in catalog photography where it was vitally important that the color used in the catalog must match precisely the color that was photographed, and it was very good for that. Digitals are theoretically more flexible and thus have the potential for better color fidelity, but to do that you basically need to know precisely which lights you are using and at that point you're in the same boat as film users. Now, is all this really compelling reasons to go to film? Not really. But I spend all day every day working in front of a computer and I'll be damned if I'm going to spend my hobby time calibrating my monitors and messing with scanners and printers (each in a dead heat for most miserable fucking peripherial ever), so I do traditional darkroom stuff, occasionally including some color work.