Optical low-pass filtering

[Bob Burman]

Confused.. help me out here a little......

 

I've understood that some of the digital cinematography cameras employ an optical low-pass filter just before the imaging sensor.

 

Now, am I wrong or is this just a euphemism for cutting the resolution in half? Much in the way that audio systems with 44100 Hz sampling freq are filtered to record only up to 22050 Hz to avoid introducing nonexistent frequencies... the whole Nyquist-theorem thingy and all that?

 

And does this mean that the vertical resolving power of a HD system is actually 1080/2=540 lines? Or do they not have low-pass filters? Or am I just plain stupid? Two beer or not two beer?

 

Bob.

[Kevin Zanit]

I believe it is just a piece of glass that cuts the UV light before hitting the censor.

 

I could be completely wrong, but I remember being told it was something along those lines (maybe not UV light, but a different frequency?)

 

 

Kevin Zanit

[Bob Burman]

Hm, I really really think it has got to do with spatial frequency i.e. resolving power.... anyone?

 

Bob.

[Kevin Zanit]

Courtesy of Elhanan (from Band-Pro?s service department):

 

"Its a piece of glass that has slits in it, it is the first piece of glass you can see after the lens mount . . . it filters out UV light . . . "

 

"It has not affect on resolution what so ever, in fact you can take it out easily yourself if you want to see what happens (not recommended though)"

 

 

Kevin Zanit

Edited March 2, 2005 by Kevin_Zanit

[David Mullen ASC]

I thought it DID roll off high-frequency detail to reduce aliasing artifacts.

[Kevin Zanit]

"It does that too" :D

[Elhanan Matos]

What it does is blur the lines between the pixels on the imager to prevent moiring(sp?), and aliasing. I believe it also has a special coating that blocks out UV light.

[Ilmari Reitmaa]

"The optical prefilter reduces the aliasing created by the periodic arrangement of CCD photosites at 8.4um pitch. It is a layered birefringent crystal that splits each light ray into four rays, each offset 8.4um vertically and horizontally on the focal plane. " [image Resolution of the One-CCD Palomar Motion Picture Camera, Charles Smith et al., 2003]

 

In effect, the low-pass filter disperses the light that would be falling on one pixel over adjacent pixels, thus creating an overlap and performing spatial frequency low-pass filtering. So actually not only does it blur the lines between the pixels, which usually cannot be resolved by lenses anyway, it convolutes adjacent pixels. So, yes, you do lose detail, however asking for any more detail would be a tradeoff. Don't know about the UV-filtering, though.

Edited March 2, 2005 by Ilmari Reitmaa

[Sam Wells]

You certainly want to block IR, or is that a separate filter ?

 

-Sam

[Elhanan Matos]

Thats it! IR not UV! The optical low pass filter also filters out Infra Red light, NOT Ultra-Violet.

[John Pytlak RIP]

Any sensor or digital display with a fixed pattern of pixels is prone to "aliasing" (usually seen as a moire pattern) when the detail in the image approaches the pixel spacing. We've all seen the moire pattern on television when the image has fine detail like a herringbone pattern suit, striped tie, or closely spaced slats on venetian blinds. Here is information about the "Nyquist Criterion" and the need for "anti-aliasing" filtration to make sure the image detail is not sharp enough to cause aliasing:

 

http://members.aol.com/ajaynejr/vidres2.htm

 

http://biology.ncsa.uiuc.edu/library/SGI_b..._html/ch02.html

 

http://www.answers.com/topic/sampling-frequency

 

http://ccrma.stanford.edu/CCRMA/Courses/25...ors/node21.html

 

http://www.stefan.winkler.net/Publications/kluwer2001.pdf

 

Because the silver halide grain "sensors" in film are random in size and distribution, film images are not prone to aliasing artifacts (unless the film is scanned with a fixed array sensor or displayed on a fixed array digital projector).

[Luke Prendergast]

IR is longer wavelength/lower frequency than visible light, so to cut infra red a high-pass filter is used. Not the same thing.

[John Sprung]

Because the silver halide grain "sensors" in film are random in size and distribution, film images are not prone to aliasing artifacts

<{POST_SNAPBACK}>

Film grain actually can alias against random picture detail. The lucky thing is that it's never irritating. In fact, it's hard even to find it even if you're looking. The one case I remember is the leaves of distant trees seeming to flutter in the wind, even though we shot on a very hot dead calm day.

 

The optical low pass has to make the Nyquist limit for the sensor array, not for the recording format. For instance, if you have a chip that's 2160 pixels high, the low pass has to be all the way out by half that, 1080 high. In optical filters, to get to zero by N/2, the rolloff has to start at N/4, or 540 lines. That's because we don't have anti-photons available, the filter has to work with positive coefficients only.

 

Downconverting from the 2160 line sensor to the 1080 line recording format is another separate resampling step, with its own Nyquist limit. But the filter for this is digital, and can come a lot closer to the optimal brick wall at N/2. That's why oversampling can get you a lot more top octave detail.

 

 

 

-- J.S.