Monday, September 23, 2013

IR on Aliasing Artifacts

Imaging Resource posted an interesting note on the recent trend to remove optical low pass filters over the sensors:

"There's been a strong move in the camera industry lately to remove low-pass filters (aka anti-aliasing filters or LPFs) from cameras, in pursuit of greater image sharpness... At IR, we feel strongly that eliminating low-pass filters is a bad idea, and a mistake for the industry. While the vast majority of natural subjects aren't subject to aliasing and moiré issues, many man-made objects have the sort of regular patterns that trigger the problem."

Imaging Resource supports its claim by real-world examples shot by one of the recent cameras with no optical anti-aliasing filter:

Example of color aliasing, caused by the fine thread patterns in the model's outfit.
Example of luminance moiré in the form of the swirly lines
in what should be diagonal louvers on the building's front.

16 comments:

  1. I do not understand the problem, as long the sensor has lower resolution than the lens there are a need for AA-filter

    ReplyDelete
    Replies
    1. The main issue is for frequencies between Nyquist/2 and Nyquist, since they are not captured by a Bayer sensor, they may introduce demoisaicing artifacts.

      Delete
    2. Increasing pixel density beyond the diffraction limit/optical detail of the lens will eliminate the problem without reducing detail.

      Regarding Foveon, bayer will do better (ISO, DR) without aliasing problems when pixel density is high enough. Bandwith and storage requirements may be reduced by combining data from several pixels before sending data of the chip if high speed is desired.

      Delete
    3. Increase pixel density beyond the diffraction limit/optical detail in stead of using AA filters will preserve the detail level from the lens.

      If lower bandwith or storage requirements are desired just combine pixel data on chip.

      Delete
  2. Buy a Sigma camera. No AA filter needed.

    Talk to me if you need a large quantity of custom chips with layered photodiodes.

    ReplyDelete
  3. Is the removal of the AA filter a clever way of reducing production costs while selling it as an advantage to the photographer?

    ReplyDelete
    Replies
    1. I'm not sure it makes a significant cost reduction for a DSLR, but some low-end Canon do have a weaker (or no) AA filter than the same resolution higher-end ones, probably because they're supposed to have lower-end lenses on them too in which case it is an advantage.

      Delete
    2. Question Mr Hornung: how would a 'weaker' OLPF reduce the cost?

      Delete
    3. Mr Anonymous: I didn't state that a weaker AA filter would reduce any cost, I was just pointing that cost is probably not the reason for making cameras prone to Aliasing.

      Delete
    4. Understood. When you said a 'weaker' AA filter, can you please be more specific? As far as I understand, all of them are four-point separation filters. If you have that information, can you please be more specific by stating distances between points relatively to the pixel dimensions for low-end camera vs. high-end camera?

      Delete
    5. Same lens on different camera bodies with very "similar" sensors (ie same brand, size, pixel pitch, noise, sensitivity, ...) lead to a different system MTF. Anybody can do the test Canon 1100D vs 400D, and 1100D vs 450D.

      Delete
    6. I thought that you were operating with more solid data when you made the initial claim. My humbled experience is that using a system MTF for an estimate of OLPF frequency response is a bit of a dubious proposal.

      Delete
    7. The AA filter has a clear impact on the system MTF, hence the initial Imaging Resource article and this post.
      I stand by my claim, which is backed by solid data, any skeptical person is welcome to acquire the same data by himself, it does not require that much equipment and is 100% non destructive black box testing of cameras, though it requires a fair amount of work. I'd be happy to provide guidance to anyone who wants, in order to avoid too much trial and error.
      Measuring a camera AA filter MTF is another story, please read my comments, all I said is that some Canon lower end cameras have a weaker AA filter than the equivalent higher-end ones, just disregard the comment if it is of no interest.

      Delete
    8. Mr Hornung, I have never doubted your ability to measure a system MTF. BTW, thank you for your offer for guidance in that matter. I still have some hesitation in accepting results of AA filter characterization using a system MTF. For example, how do you discriminate between effects of AA, sensor MTF or signal processing onto system MTF in Canon cameras?
      Anyway, what really caught my attention is your statement that some Canon cameras have a weaker (OR NO) AA filter. I haven’t found yet a Canon camera without an AA filter. Having said that, I dissected only half a dozen of prosumer and consumer cameras so I would appreciate it if you could tell me which Canon camera doesn’t have an AA filter.

      Delete
  4. There are ways (at least clues) to find out if RAW has been processed or not, and as far as I know (haven't look at the latest models though) Canon RAW is unprocessed except for defective pixels.
    I don't think that any Canon DSLR has no AA filter either, but don't rule it out as a possible "weaker" case (hence the OR statement).
    Actually there is a way to check how much the AA filter affects system MTF, using a simple polarizing filter.
    Do you have any data/picture of your dissections online?

    ReplyDelete
    Replies
    1. I do not have a web page, but I can share with you with some examples. Some of the examples will demonstrate limitation of the polarizing filter method. Please contact me at olpfinfo314@gmail.com

      Delete

All comments are moderated to avoid spam.