11-26-2010, 09:50 AM
[quote name='Klaus' timestamp='1290753217' post='4452']
1. the strength of an AA filter is a matter of taste. There's no wrong or right here.
[/quote]
AA filter strength is certainly not a matter of taste. Too high and details is lost unnecessarily. Too low and the sensor will record false detail. The latter is not a problem only if the input signal is sufficiently bandwidth limited anyhow. For instance, a weak lens or suboptimal focus can achieve this. A good lens focused on a high frequency pattern will however create a signal above the Nyquist frequency and what you record then is the result of aliasing, i.e., signal above the frequency threshold that the sensor can capture mirrored into the signal band. In other words, rubbish. I cannot tell you what made Leica or Kodak come to the conclusion that colour moiré is better for their users than appropriate capture sharpening. Maybe their logic is that is more important to convince pixel peepers than to avoid colour moiré at all cost.
Admittedly, around the optimal strength of an AA filter there is an admissible range and where you want to sit in that range is an engineering decision. However, I dispute the notion of "taste".
[quote name='Klaus' timestamp='1290753217' post='4452']
2. there're many cameras with weak AA filters (A33, NEX-5, all MFTs, D7000, K10D) and there're next to zero user complaints about this. That's a fact and no theory.
[/quote]
The lack of user complaints doesn't justify questionable engineering decisions. If something is wrong, it doesn't become right because most don't notice. Again, any detail a sensor records in the absence of an AA filter, it should never record in the first place.
[quote name='Klaus' timestamp='1290753217' post='4452']
3. yes, a weak AA filter causes moirees in certain situations - that's not a theoretical aspect but a practical one. However, this is often blown way out-of-proportions (-> (2)). Markus showed me some converted (Leica M) RAWs via CaptureOne. C1 seems to have a fairly efficient anti-moiree algorithm (unlike Silkypix) and CaptureOne is one of the sharpest RAW converters out there. It is not able to eliminate the problem altogether of course. It is the question whether this is an issue for your personal photography. It may be fore yours and your taste.
[/quote]
Fact is, as you say, that moiré represents damage to the image that can never be fully removed (not without losing detail anyhow). The detail lost by an AA filter, however, is fully recoverable through capture sharpening. That's the idea of respecting the Nyquist frequency (i.e., sampling with the Nyquiste rate).
[quote name='Klaus' timestamp='1290753217' post='4452']
4. the K5 has a strong AA filter. I don't like it personally and it is counterproductive for the lens tests. I don't want to measure blur after all. I don't test the camera here - I test lenses.
[/quote]
So do you disagree with Falk's assessment that the K-5's AA filter is "just right"? I don't agree that it is "strong". Now I understand that your tests are a special case because you are working with B&W patterns only, aren't you? This means that for your application a B&W sensor would be ideal. In other words, the AA filter strength that is needed for a colour Bayer matrix sensor is too strong for your application. Could you please go into a bit of detail, though, how you process your RAW files? Clearly, a standard processing would not be optimal as you must somehow built in the assumption of a pure B&W signal. Also you must somehow prevent the analysis to incorrectly interpret aliasing artifacts (in the case of the K10D). I'm genuinly interested in your workflow and how it justifies that the K-5 isn't more useful to you.
[quote name='Klaus' timestamp='1290753217' post='4452']
5. the difference between the K5 and D7000 is certainly higher than just 4% in RAW mode. It is higher than 4% vs the A33 already.
[/quote]
So GordonBGood is wrong?
[quote name='Klaus' timestamp='1290753217' post='4452']
6. at 16mp many lenses start to act as a low pass filter because they don't reach the sensor performance (due to the high pixel density digicams had no AA filter since the dawn of time). That's certainly true regarding the borders performance of most lenses.
[/quote]
Good lenses will excite moiré in a 16MP sensor with an AA filter that is too weak. One can see that in the IR test shots. A very good lens will peak at f/5.6 which means it requires 27MP for green light. Of course there are even better lenses. I don't think in this context we are talking about lens border performance.
[quote name='Klaus' timestamp='1290753217' post='4452']
7. the K10D has no vertical filtering. It can reach resolution figures well beyond Nyquist here.
[/quote]
What you are saying is that with N pixels you can count more than N/2 lines. That's not possible. There is a way to make sense of what you said but I wonder whether you agree with Falk on the explanation. You seem to disagree with him on other accounts, so I'd be interested in how you personally justify your above statement.
Thanks a lot for responding. I would have appreciated more technical explanation as to how you arrive at your K10D -> 2350 LW/PH and K-5 -> 2500 LW/PH statement. Certainly such a statement should always be accompanied with the special conditions (B&W signal only) for the K10D that are required to make it half plausible. Certainly for colour photography this comparison is flawed and I don't follow your negative comments about the strength of the K-5's AA filter in a general context.
If you could provide us with more information about the assumptions you are making and/or the workflow you are using, it might help us to understand why you seem to be in disagreement with other experts (GordonBGood and Falk Lumo to name two).
1. the strength of an AA filter is a matter of taste. There's no wrong or right here.
[/quote]
AA filter strength is certainly not a matter of taste. Too high and details is lost unnecessarily. Too low and the sensor will record false detail. The latter is not a problem only if the input signal is sufficiently bandwidth limited anyhow. For instance, a weak lens or suboptimal focus can achieve this. A good lens focused on a high frequency pattern will however create a signal above the Nyquist frequency and what you record then is the result of aliasing, i.e., signal above the frequency threshold that the sensor can capture mirrored into the signal band. In other words, rubbish. I cannot tell you what made Leica or Kodak come to the conclusion that colour moiré is better for their users than appropriate capture sharpening. Maybe their logic is that is more important to convince pixel peepers than to avoid colour moiré at all cost.
Admittedly, around the optimal strength of an AA filter there is an admissible range and where you want to sit in that range is an engineering decision. However, I dispute the notion of "taste".
[quote name='Klaus' timestamp='1290753217' post='4452']
2. there're many cameras with weak AA filters (A33, NEX-5, all MFTs, D7000, K10D) and there're next to zero user complaints about this. That's a fact and no theory.
[/quote]
The lack of user complaints doesn't justify questionable engineering decisions. If something is wrong, it doesn't become right because most don't notice. Again, any detail a sensor records in the absence of an AA filter, it should never record in the first place.
[quote name='Klaus' timestamp='1290753217' post='4452']
3. yes, a weak AA filter causes moirees in certain situations - that's not a theoretical aspect but a practical one. However, this is often blown way out-of-proportions (-> (2)). Markus showed me some converted (Leica M) RAWs via CaptureOne. C1 seems to have a fairly efficient anti-moiree algorithm (unlike Silkypix) and CaptureOne is one of the sharpest RAW converters out there. It is not able to eliminate the problem altogether of course. It is the question whether this is an issue for your personal photography. It may be fore yours and your taste.
[/quote]
Fact is, as you say, that moiré represents damage to the image that can never be fully removed (not without losing detail anyhow). The detail lost by an AA filter, however, is fully recoverable through capture sharpening. That's the idea of respecting the Nyquist frequency (i.e., sampling with the Nyquiste rate).
[quote name='Klaus' timestamp='1290753217' post='4452']
4. the K5 has a strong AA filter. I don't like it personally and it is counterproductive for the lens tests. I don't want to measure blur after all. I don't test the camera here - I test lenses.
[/quote]
So do you disagree with Falk's assessment that the K-5's AA filter is "just right"? I don't agree that it is "strong". Now I understand that your tests are a special case because you are working with B&W patterns only, aren't you? This means that for your application a B&W sensor would be ideal. In other words, the AA filter strength that is needed for a colour Bayer matrix sensor is too strong for your application. Could you please go into a bit of detail, though, how you process your RAW files? Clearly, a standard processing would not be optimal as you must somehow built in the assumption of a pure B&W signal. Also you must somehow prevent the analysis to incorrectly interpret aliasing artifacts (in the case of the K10D). I'm genuinly interested in your workflow and how it justifies that the K-5 isn't more useful to you.
[quote name='Klaus' timestamp='1290753217' post='4452']
5. the difference between the K5 and D7000 is certainly higher than just 4% in RAW mode. It is higher than 4% vs the A33 already.
[/quote]
So GordonBGood is wrong?
[quote name='Klaus' timestamp='1290753217' post='4452']
6. at 16mp many lenses start to act as a low pass filter because they don't reach the sensor performance (due to the high pixel density digicams had no AA filter since the dawn of time). That's certainly true regarding the borders performance of most lenses.
[/quote]
Good lenses will excite moiré in a 16MP sensor with an AA filter that is too weak. One can see that in the IR test shots. A very good lens will peak at f/5.6 which means it requires 27MP for green light. Of course there are even better lenses. I don't think in this context we are talking about lens border performance.
[quote name='Klaus' timestamp='1290753217' post='4452']
7. the K10D has no vertical filtering. It can reach resolution figures well beyond Nyquist here.
[/quote]
What you are saying is that with N pixels you can count more than N/2 lines. That's not possible. There is a way to make sense of what you said but I wonder whether you agree with Falk on the explanation. You seem to disagree with him on other accounts, so I'd be interested in how you personally justify your above statement.
Thanks a lot for responding. I would have appreciated more technical explanation as to how you arrive at your K10D -> 2350 LW/PH and K-5 -> 2500 LW/PH statement. Certainly such a statement should always be accompanied with the special conditions (B&W signal only) for the K10D that are required to make it half plausible. Certainly for colour photography this comparison is flawed and I don't follow your negative comments about the strength of the K-5's AA filter in a general context.
If you could provide us with more information about the assumptions you are making and/or the workflow you are using, it might help us to understand why you seem to be in disagreement with other experts (GordonBGood and Falk Lumo to name two).