[Brightcolours]

[quote name='popo' timestamp='1305909208' post='8692']

In the case of bayer sensor, wouldn't the "strength" of AA filter be some two photosites of blur to allow for the sparse "chroma" detectors to work effectively. Foveon would need about one photosite strength. And before you go there, I know you can still extract some RGB info from all photosites, but let's say the "non-native" colour data is limited in value.

[/quote]

The AA-filter has NOTHING to do with what colour light is being captured right under it.

All the AA-filter ideally does, is make sure that the image is blurred enough, so that the resulting image does not have frequencies exceeding the sample frequency. In case of the sensors, the sampling frequency is related to the pixel pitch.

And that then is EXACTLY the same, for a foveon style sensor and a Bayer CFA style sensor.



Again, detail is in the luminance, not in the chrominance, There are no "gaps" in the luminance capturing with a Bayer CFA sensor.

Detail wise, no real difference, between foveon and Bayer CFA. But you can't compare them, because foveon doesn't put an AA-filter on it.

[quote name='popo' timestamp='1305909208' post='8692']

I'd consider moire in this case a manifestation of aliasing, so I wouldn't consider that a separate category, but more a subset.

[/quote]

Aliasing is you seeing edges where there are no edges, usually identified by "stair case" lines/edges. Moire patterns are the patterns that appear when sampling a higher frequency than the sensor can handle.

Both have to do with the arbitrary edges of each pixel, which do not necessarily correspond with the scene/projected image, but you can't call moire a kind of aliasing. Better is to say that it is CAUSED by aliasing.



[quote name='popo' timestamp='1305909208' post='8692']

Side note: when I converted the 300D to IR by removing the existing filter assembly, I didn't realise at the time I also removed the AA filter which was in the same module. I thought it looked sharper afterwards. Very rarely do I see something resembling moire or aliasing on it, although the longer wavelengths of IR do cause more diffraction blurring so may offset that itself.





I agree there, but that is the way of marketing.

[/quote]

Of course it looked sharper, sharp edges always look sharper. But the edges shown by the square pixels do not necessarily correspond to the edges in the scene... pixels only can have one value per pixel, no matter what gradations actually happen in the scene below that pixel. So... you feel you see more detail, but it is (almost always) false detail.



Preference is entirely personal.

[popo]

[quote name='Brightcolours' timestamp='1305911637' post='8694']

Again, detail is in the luminance, not in the chrominance, There are no "gaps" in the luminance capturing with a Bayer CFA sensor.

[/quote]

Only half the photosites are nominally luminance ones in that model.

Aliasing is you seeing edges where there are no edges, usually identified by "stair case" lines/edges. Moire patterns are the patterns that appear when sampling a higher frequency than the sensor can handle.

I'm thinking in the sampling theory definition of aliasing - the recording of frequency domain data outside that which is desired. I favour the use of oversampling to overcome that, which in sensor terms means more MP density.

you can't call moire a kind of aliasing. Better is to say that it is CAUSED by aliasing.

I thought I did say that, well, manifestation as a longer way round to saying caused. Subset in the sense it is one way aliasing can appear.

[Brightcolours]

[quote name='popo' timestamp='1305914787' post='8695']

Only half the photosites are nominally luminance ones in that model.



[/quote]

No, they are not. The red and blue sensels also deliver luminance information.

[popo]

[quote name='Brightcolours' timestamp='1305916015' post='8696']

No, they are not. The red and blue sensels also deliver luminance information.

[/quote]

I acknowledged that earlier, hence the use of the term "nominally" here. You can only recover that information in conjunction with values from neighbouring photosites. The data across those photosites needs to be sufficiently consistent for this to work well. An AA filter appropriate to foveon would not be for bayer pattern of same sensor pitch.

[Brightcolours]

[quote name='popo' timestamp='1305921134' post='8700']

I acknowledged that earlier, hence the use of the term "nominally" here. You can only recover that information in conjunction with values from neighbouring photosites. The data across those photosites needs to be sufficiently consistent for this to work well. An AA filter appropriate to foveon would not be for bayer pattern of same sensor pitch.

[/quote]

No, that is not correct. The luminance information is more detailed than you seem to think.



It is still the pixel pitch which determines the optimal AA-filter, no matter what colour system lays beneath.

[anyscreenamewilldo]

some nice samples . . . but i reckon the 100-300 i have would give the comparison 120-300 shot a good run for its money - except at 2.8

[popo]

[quote name='Brightcolours' timestamp='1305924614' post='8704']

No, that is not correct. The luminance information is more detailed than you seem to think.



It is still the pixel pitch which determines the optimal AA-filter, no matter what colour system lays beneath.

[/quote]

I think we'll have to agree to disagree on this point, unless either of us bothers to find some credible references to support the respective cases.

[Guest]

[quote name='popo' timestamp='1305914787' post='8695']

Only half the photosites are nominally luminance ones in that model.[/quote]

Actually all photosites provide luminance information, i.e. pure black and white pattern will look the same for R, G or B photosite. However, going back to old dispute with Brightcolours, there are situations where Foveon's approach does have an edge over Bayer's - namely where detail is created by the change in light intensity for specific wavelength only. For instance, red or blue flower petals, etc. The biggest nightmare for Bayer sensor would be black pattern on pure red background - G and B sensors would see it as black (red wavelength blocked) and effective resolution would be reduced by 4 <img src='http://forum.photozone.de/public/style_emoticons/<#EMO_DIR#>/tongue.gif' class='bbc_emoticon' alt='' />

[Brightcolours]

[quote name='Lomskij' timestamp='1305936900' post='8710']

Actually all photosites provide luminance information, i.e. pure black and white pattern will look the same for R, G or B photosite. However, going back to old dispute with Brightcolours, there are situations where Foveon's approach does have an edge over Bayer's - namely where detail is created by the change in light intensity for specific wavelength only. For instance, red or blue flower petals, etc. The biggest nightmare for Bayer sensor would be black pattern on pure red background - G and B sensors would see it as black (red wavelength blocked) and effective resolution would be reduced by 4 <img src='http://forum.photozone.de/public/style_emoticons/<#EMO_DIR#>/tongue.gif' class='bbc_emoticon' alt='' />

[/quote]

That could only happen if the CFA actually would just let through one wavelength, which is doesn't. Each "colour" lets through a wider band of the spectrum, making your example more or less purely theoretical...

[popo]

[quote name='Lomskij' timestamp='1305936900' post='8710']

Actually all photosites provide luminance information[/quote]



Since this is *still* being bought up, again I reiterate I acknowledge that *some* luminance information can be gathered by the others, but their value is significantly lower in that role.



Also, for fun let's read the original Bayer pattern "Color imaging array" Patent. At least I would in more detail if I wasn't leaving for a shoot in a moment.