Tuesday, October 22, 2013

Pentax DSLR Features "Anti-Aliasing Filter Simulator"

A newly announced Pentax K-3 DSLR features "the world’s first AA filter simulator, which reproduces the effects created by an optical AA filter. By applying microscopic vibrations to the CMOS sensor during exposure, the K-3 minimizes false color and moiré. You have a choice of three settings to obtain the desired effect: “TYPE 1” to attain the optimum balance between image resolution and moiré; “TYPE 2” to prioritize moiré compensation; and “OFF” to prioritize image resolution. Thanks to this innovative feature, the K-3 offers the benefits of two completely different cameras — the high-resolution images assured by an AA-filter-free model, and minimized false color and moiré assured by an AA-filter-equipped one. You can switch the AA filter effect on and off as you wish."

A Youtube video shows how it works:


  1. Dear Mr. Pentax, Prof. Reimar Lenz will love you !!!! See US2003011747.

    1. This is abandoned application. There is another one by Kodak WO2008079187 that appears to be abandoned too.

    2. I worked on a similar concept 20 years ago. At that time, I made a pixel which can integrate photodiode in 2 directions during the X-Y movement. By doing so, we can make image convolution with any kernel in real-time ... By the problem is that when you want to do at high speed, the vibrator will consume a huge power :( !!

      In this proposal, if the exposure time is short, it should be hard to move the sensor !! At low speed, there is no need. So not cool !

      -yang ni

  2. This does not work correctly with moving objects. Il the moving object and the sensor are moving at the same speed, the moiré effect is still present. A linear movement is enough to show moire on on axis.

  3. A motion-blurred image doesn't produce moire.
    And I can't think of any target that could possibly match the movement of the sensor.

  4. In my company, we have set such a system up. You get moire when :
    - an object moves synchronously with the sensor
    - the camera does a traveling synchronously with the sensor.

    We tough these are very special conditions quite impossible to get in real life but we were wrong.

  5. More details :
    Moving synchronously does'nt mean moving like the sensor. If the sensor describes a circle, the object does'nt have to describe a circle as well. Just think of one axis.
    If you shoot a car moving from left to right, then :
    - when the sensor moves from right to left, you get a blurred picture
    - when the sensor moves from left to right, you get a non blurred picture with moire if the car moves at the same speed than the sensor.
    The resulting picture is the addition of the blurred picture and the moire picture. Finaly, you see only the moire artifacts.

  6. But the sensor do many cycles in one exposure, doesn't it? And photographed object is usually not vibrating ...

  7. To start with, experience with the previous 16mp Pentax K5IIs, which has no anti-aliasing filter, is that moiré is very rare in practice. It will be even rarer at the 24mp resolution of the K3. Then you are thinking about a very rare coincidence of motion between the subject and the sensor. So, is the Pentax model foolproof? Probably not. Is it useful? Absolutely! The more practical limitation of the SR filter is that its effectiveness is reduced for exposures shorter than 1/1000 sec, because then the sensor does not have time to do its complete pattern of motion. But I can live with that.

  8. Let simplify the problem. Forget the Y axis and focus on the X axis. So, imagine the sensor is "vibrating" only left-right.
    With a moving car, going from left to right :
    - when the sensor moves from right to left, you get a blurred car picture because from the sensor point of view, the car is moving. You also get a blurred picture for all static elements in the picture because the sensor is moving.
    - when the sensor is moving from left to right, all static elements of the picture are blurred by the sensor movement. But, if the car is moving at the same speed the sensor moves, the car is NOT blurred because from the sensor point of view, the car is static. Then you get moire.
    You have the same phenomenon on the Y axis of course.
    Now, imagine a tree with thousands of leaves in the wind...

    1. But the sensor does move in X and Y, and the target can only match one of the four movements, so you end up with, worst case, a mix of 4 images, 3 of them nicely blurred, one of them with moire. Not the strongest OLPF in the world, but that's going to be good enough for most people.
      This problem is a detail (1/4 of the original problem) for an exception (the target moving in sync with the sensor) of an exception (the target actually being able to generate moire on this sensor).

    2. The fact that the sensor moves X and Y does'nt change anything to the problem. This is not a mix of 4 images. The sensor is moving continuously. Generaly, the sensor will describe a circle because other paterns are noisy and energy hungry. If you shoot a patern of high frequency white and black bars, when blurred, you get grey. Something very neutral. But when these bars generate moire, you get color bars, Something realy not neutral. Adding grey to colors gives colors.

  9. Funny, only a few years ago the sensor was moved to *increase* the resolution (see Jitter camera or the Leica DFC550). I am also trying to wrap my had around what this will do to motion induced rolling shutter artifacts.



All comments are moderated to avoid spam.