10-01-2010, 05:51 AM
[quote name='Brightcolours' timestamp='1285847839' post='3390']
I really fail to see how diffraction softening could happen on pixel level, they are using a wrong term to describe something different.
[/quote]
Think of it as a single slit diffraction experiment... the slit ofcourse is the Foveon pixel opening at the front. Fitting 4800 pixels on a sensor 24mm wide means a pixel is theoretically 5000nm wide. However, Foveon sensors have a fill factor of around 40%. If we round it up to 50%, the width of the slit (pixel opening) is 2500nm wide. This is a bit wider than the longest visible red wavelengths (which are around [url="http://en.wikipedia.org/wiki/Visible_spectrum#Spectral_colors"]700nm[/url] but it's quite close. But I'm giving the benefit of the doubt to Sigma here though... they wouldn't announce the sensor if such a fundamental problem was there.
[quote name='Brightcolours' timestamp='1285847839' post='3390']
Point remains: The foveon sensors will not be affected worse by diffraction. Diffraction is ONLY in the projected image, and the AA-filter-less sampling willnot get a more soft picture than an AA-filtered one.
[/quote]
What is the meaning of soft? Is it when comparing a 14MP Bayer CFA image vs. a 14MP Foveon? You need to establish the pixel sizes first before talking about the effects of diffraction.
[quote name='Brightcolours' timestamp='1285847839' post='3390']
Which brings us back to the reason Sigma images are "sharper"... due to the refusal to use AA-filters. If you remove the AA-filter from a Bayer CFA sensor, you will get the same kind of sharpness. Liked by some, disliked by those in the know (that is why most camera manufacturers put AA-filters in front).
[/quote]
As a person who has a 50D without an AA-filter and lenses that can out-resolve a 50D without an AA-filter, I can say that this is not true. Plain removal of the AA-filter doesn't give the same resolution of a Foveon (again, what sizes are the compared sensors here?). In fact, from my experience, when the AA-filter of a Bayer CFA is properly matched to the pixel sizes of the sensor, that is when the produced images are more "Foveon-like". Removing the AA-filter only helps if it's overly strong, such as in the case of the D3/D700 (i.e. when compared to the original 5D).
Also removing the AA-filter (and when the lenses are not the limiting factor) will lead to funny demosaic'ing artifacts. This is not at all Foveon-like. It might fool the eye in to thinking that it's real detail, but it looks really ugly at 100%. Just look at a full size sharp Leica M9 image for example. With or without the AA-filter, a Bayer CFA still needs to combine roughly 3 adjacent photosite values in to 1 to produce 1 RGB pixel and this is the actual size of a Bayer "pixel" (i.e. 1/3 of the total photosite count). Foveon does this with 1. So Foveon can have a fewer larger number of pixels and get the same resolution.
Don't take my word for it... take a look here at the link below. This guy modified a 4.7MPixel Sigma SD14 to take EOS lenses and compared it against a 50D (15MPhotosites). The lens he used on both was the Canon 200 f/1.8L @ f/4 (i.e. very very sharp). The images are at their native resolutions, but if you upscale the SD14 image (e.g. bridge image) to match the dimensions of the corresponding 50D image, using a good upscaling algorithm, you'll see that it's difficult to tell them apart just by looking at the detail:
http://www.pbase.com/bigflat/sd14_50d
So that's the relationship between the Bayer CFA and Foveon designs... removing the AA filter doesn't equalise them.
I really fail to see how diffraction softening could happen on pixel level, they are using a wrong term to describe something different.
[/quote]
Think of it as a single slit diffraction experiment... the slit ofcourse is the Foveon pixel opening at the front. Fitting 4800 pixels on a sensor 24mm wide means a pixel is theoretically 5000nm wide. However, Foveon sensors have a fill factor of around 40%. If we round it up to 50%, the width of the slit (pixel opening) is 2500nm wide. This is a bit wider than the longest visible red wavelengths (which are around [url="http://en.wikipedia.org/wiki/Visible_spectrum#Spectral_colors"]700nm[/url] but it's quite close. But I'm giving the benefit of the doubt to Sigma here though... they wouldn't announce the sensor if such a fundamental problem was there.
[quote name='Brightcolours' timestamp='1285847839' post='3390']
Point remains: The foveon sensors will not be affected worse by diffraction. Diffraction is ONLY in the projected image, and the AA-filter-less sampling willnot get a more soft picture than an AA-filtered one.
[/quote]
What is the meaning of soft? Is it when comparing a 14MP Bayer CFA image vs. a 14MP Foveon? You need to establish the pixel sizes first before talking about the effects of diffraction.
[quote name='Brightcolours' timestamp='1285847839' post='3390']
Which brings us back to the reason Sigma images are "sharper"... due to the refusal to use AA-filters. If you remove the AA-filter from a Bayer CFA sensor, you will get the same kind of sharpness. Liked by some, disliked by those in the know (that is why most camera manufacturers put AA-filters in front).
[/quote]
As a person who has a 50D without an AA-filter and lenses that can out-resolve a 50D without an AA-filter, I can say that this is not true. Plain removal of the AA-filter doesn't give the same resolution of a Foveon (again, what sizes are the compared sensors here?). In fact, from my experience, when the AA-filter of a Bayer CFA is properly matched to the pixel sizes of the sensor, that is when the produced images are more "Foveon-like". Removing the AA-filter only helps if it's overly strong, such as in the case of the D3/D700 (i.e. when compared to the original 5D).
Also removing the AA-filter (and when the lenses are not the limiting factor) will lead to funny demosaic'ing artifacts. This is not at all Foveon-like. It might fool the eye in to thinking that it's real detail, but it looks really ugly at 100%. Just look at a full size sharp Leica M9 image for example. With or without the AA-filter, a Bayer CFA still needs to combine roughly 3 adjacent photosite values in to 1 to produce 1 RGB pixel and this is the actual size of a Bayer "pixel" (i.e. 1/3 of the total photosite count). Foveon does this with 1. So Foveon can have a fewer larger number of pixels and get the same resolution.
Don't take my word for it... take a look here at the link below. This guy modified a 4.7MPixel Sigma SD14 to take EOS lenses and compared it against a 50D (15MPhotosites). The lens he used on both was the Canon 200 f/1.8L @ f/4 (i.e. very very sharp). The images are at their native resolutions, but if you upscale the SD14 image (e.g. bridge image) to match the dimensions of the corresponding 50D image, using a good upscaling algorithm, you'll see that it's difficult to tell them apart just by looking at the detail:
http://www.pbase.com/bigflat/sd14_50d
So that's the relationship between the Bayer CFA and Foveon designs... removing the AA filter doesn't equalise them.