12-07-2016, 12:26 AM
The tests were NEVER cross-comparable - we mention this numerous times throughout the site. They are more or less comparable between cameras of similar resolution. If a review site using a chart-based test system states differently, they are plain lying to you.
Honestly I was never comfortable with the numeric LW/PH values because people tend to over-interpret those. These figures have always been "virtual". Again, this applies to all review sites.
Anyway, what we are analysing are images that resemble those produced by users. Or in other words - the images are ALWAYS sharpened to some degree. We tune the sharpening (unsharp masking) to a neutral value provided by the imatest tool. This translates comparatively mild sharpening (less than straight camera JPEGs).
In essence this is similar to the discussion about AA filters - AA filters behave like reverse unsharp masking. So if you have two cameras with the same -say- 36mp sensor but different AA filters, the output is different. Further impacts come from micro-lenses, A/D converters and in-camera RAW pre-processing and external RAW processing (bayer interpolation). Point is - the output will always be different thus whatever you do - the base images are already not cross-comparable.
Now about the effect on those LW/PH values:
Imagine a black/white edge that is blurred across -say- 10 pixels.
Then imagine a black/white edge that is blurred across -say- 2 pixels.
What is the effect of your default sharpening in these cases ? Well, in the 1st case there's little sharpening at all because the sharpening radius is pretty much exceeded. In the 2nd case, the sharpening is in effect.
In Imatest this 2nd case is boosted in terms of LW/PH which is what you are seeing in the 21mp MTFs (relative to the 50mp MTFs).
Now if you compare a 50mp image with a 21mp image, the 21mp image will appear to be sharper ON PIXEL LEVEL because of the lower resolution. That is exactly the message of those MTF charts. So it is a statement how a lens performs relative to the underlying "sensor system".
Of course, we could provide pure figures without sharpening at all but then you wouldn't really see a connection between what you are seeing on screen and those charts. But what would be the point ? Nobody uses images without sharpening.
(Note: images from a Foveon sensor would only be marginally be sharpened because they have a neutral sharpness to start with).
If you want cross comparable results, you have to look for MTFs obtained from an optical bench (without camera). Do these figures help you? Well, in my opinion they don't because it neglects the effect from the image sensor and digital post-processing.
Honestly I was never comfortable with the numeric LW/PH values because people tend to over-interpret those. These figures have always been "virtual". Again, this applies to all review sites.
Anyway, what we are analysing are images that resemble those produced by users. Or in other words - the images are ALWAYS sharpened to some degree. We tune the sharpening (unsharp masking) to a neutral value provided by the imatest tool. This translates comparatively mild sharpening (less than straight camera JPEGs).
In essence this is similar to the discussion about AA filters - AA filters behave like reverse unsharp masking. So if you have two cameras with the same -say- 36mp sensor but different AA filters, the output is different. Further impacts come from micro-lenses, A/D converters and in-camera RAW pre-processing and external RAW processing (bayer interpolation). Point is - the output will always be different thus whatever you do - the base images are already not cross-comparable.
Now about the effect on those LW/PH values:
Imagine a black/white edge that is blurred across -say- 10 pixels.
Then imagine a black/white edge that is blurred across -say- 2 pixels.
What is the effect of your default sharpening in these cases ? Well, in the 1st case there's little sharpening at all because the sharpening radius is pretty much exceeded. In the 2nd case, the sharpening is in effect.
In Imatest this 2nd case is boosted in terms of LW/PH which is what you are seeing in the 21mp MTFs (relative to the 50mp MTFs).
Now if you compare a 50mp image with a 21mp image, the 21mp image will appear to be sharper ON PIXEL LEVEL because of the lower resolution. That is exactly the message of those MTF charts. So it is a statement how a lens performs relative to the underlying "sensor system".
Of course, we could provide pure figures without sharpening at all but then you wouldn't really see a connection between what you are seeing on screen and those charts. But what would be the point ? Nobody uses images without sharpening.
(Note: images from a Foveon sensor would only be marginally be sharpened because they have a neutral sharpness to start with).
If you want cross comparable results, you have to look for MTFs obtained from an optical bench (without camera). Do these figures help you? Well, in my opinion they don't because it neglects the effect from the image sensor and digital post-processing.