12-02-2010, 07:51 AM
In simple words: an AA filter softens an image in order to avoid moire/aliasing effects. The softening can be (pretty much) reversed via a certain amount USM due to the linear nature of optical systems. Thus a very defined amount of USM is recovering (most of) the original amount of "resolution" information (at cost of sensor noise though). Frankly that's just what most of you do with your RAWs. Nobody uses completely unsharpened RAWs except maybe in very high ISO images where noise is more critical than "resolution".
Just to mention - we don't apply any aggressive sharpening (I strongly opposed that approach in the recent K5 discussion) anyway. However, even if we did this would have an impact on the scale only but not on the qualitative validity of the results. A sharp center would still be sharp and a soft corner would still be soft - that's just a matter of the presentation of the resulting values.
We have always mentioned that the results are technically not cross-comparable. They are not comparable because of different AA filters, different sensor resolutions and different post-processing (both internal and external). The processing chain is simply different.
As far as publishing RAWs is concerned - we are generating multiple gigabytes of data per test. There's a natural limit to the bandwidth of our service ... our budget. :-)
[quote name='Tiz' timestamp='1291274521' post='4673']
What does this mean in laymen's terms?
1. A body with a thick AA filter not only requires but also allows for more sharpening and, therefore, it would be unfair to compare it with a body that does not have an AA filter, if the test procedure is based on unsharpened files?
2. If the test procedure is based on RAW files differently sharpened for each body, the amount of sharpening is probably not exactly right and, therefore, grants an unfair advantage to one of the bodies?
3. If the test procedure is based on jpeg files, mainly the quality of the jpeg engine (and not the sensor) is tested?
If so: wouldn't it make sense to make the RAW files of the test shots available for download so that each reader can also try to develop the RAWs with his or her personally preferred RAW converter in order to figure out whether there are any relevant differences?
[/quote]
Just to mention - we don't apply any aggressive sharpening (I strongly opposed that approach in the recent K5 discussion) anyway. However, even if we did this would have an impact on the scale only but not on the qualitative validity of the results. A sharp center would still be sharp and a soft corner would still be soft - that's just a matter of the presentation of the resulting values.
We have always mentioned that the results are technically not cross-comparable. They are not comparable because of different AA filters, different sensor resolutions and different post-processing (both internal and external). The processing chain is simply different.
As far as publishing RAWs is concerned - we are generating multiple gigabytes of data per test. There's a natural limit to the bandwidth of our service ... our budget. :-)
[quote name='Tiz' timestamp='1291274521' post='4673']
What does this mean in laymen's terms?
1. A body with a thick AA filter not only requires but also allows for more sharpening and, therefore, it would be unfair to compare it with a body that does not have an AA filter, if the test procedure is based on unsharpened files?
2. If the test procedure is based on RAW files differently sharpened for each body, the amount of sharpening is probably not exactly right and, therefore, grants an unfair advantage to one of the bodies?
3. If the test procedure is based on jpeg files, mainly the quality of the jpeg engine (and not the sensor) is tested?
If so: wouldn't it make sense to make the RAW files of the test shots available for download so that each reader can also try to develop the RAWs with his or her personally preferred RAW converter in order to figure out whether there are any relevant differences?
[/quote]