Forum Twitter Facebook Google + RSS

Author Topic: Anti-aliasing filter  (Read 2312 times)

Offline kgyd

  • Newbie
  • *
  • Posts: 3
Anti-aliasing filter
« on: November 26, 2013, 12:49:42 pm »
This a comment for this page on the website
"Low-pass filter:  No AA filter, AA filter simulation"

Knowing the basic method of operation but no details about how this is implemented, I would rather say the following:

From a signal processing point of view, the method used in K-3 is effectively an oversampling. You move the sensor a subpixel distance and also measure in between the original pixel locations. That is you measure at a shorter wave length (higher frequency) than the distance (twice the distance, actually) of neighbouring pixels. This is still not an anti-aliasing (AA) filter but the final pixel size is equivalent to integration over the surface area of the pixel, which act as a low-pass filter. (In other words, the sensor is not sensitive to spatial variation of the signal over a distance shorter that the radius of the pixel.) The problem, I is that this "integrating filter" is not efficient enough at the Nyquist-frequency. The oversampling realized in K-3 is certainly very useful in this regard. The baseline is that the method indeed implements an AA filter, not just a "simulation".

In the same time, this method certainly brings sacrifice in the parameters measured by DxO Labs. It would be most interesting to see how dynamic range, color depth, and low-ISO performance are influenced by the different AA-filter settings of K-3. What is also to be considered that the very reason for removing the OLPF is to increase resolution. Resolution is not included in the sensor performance DxO Mark. In this regard, P-Mpix values with different lenses could give a hint.

To make justice to the K-3, I would list values with AA filter off in the sensor ratings and P-Mpix values.