OLPF-mod/fish-net/back-filter for Moire?

[USLatin]

The OLPF blurs, lines, both horizontal and vertical, possibly in different layers stacked in opposite directions and with different frequency cut-offs. The OLPF is there to avoid that overlap in the chart, but when it comes down to it, an OLPF simply creates a low cut-of point, for resolved lines, detail. The camera designer's job is to match the right OLPF with the right sensor...

In this case the OLPF is (wonderfully) designed for the 18MP CMOS in the 7D/T2i... unfortunately only parts of the sensor work in video mode, so the sensor becomes a completely different one and the OLPF isn't matched to it anymore. We have a camera without one of the most important parts of any professional or even just decent camera.

 

[Martti Ekstrand]

Then how come datasheets like the one I linked to claim these are UV blocking filters and make no mention of blurring?
...
If that were the case, you would LOSE that frequency and thus you would lose that specific color. Detail is not a function of frequency, it is a function of photons being resolved in a pattern. If you were to somehow reduce the wavelength(frequency) you would shift color. Does not compute.

That's because you are confounding wavelength low-pass and detail low-pass.

 

[USLatin]

That's because you are confounding wavelength low-pass and detail low-pass.

Yes. That is why. You are right.

 

[meauounji]

Update:

Camera is on its way back. Arriving tomorrow.

my breath is bated.

 

[Svart]

Well, yes, UV and IR are rejected through dispersion. That's the most efficient way of controlling those spectra, to limit their transmission through scattering (refraction/reflection/defraction/deflection).

Think of sunscreen, it works by creating a coating that scatters UV light before it hits your skin.

and yes the IR coating is just as significant because CMOS and CCD technology are both more sensitive in IR than they are in visible light. If you didn't have it, your sensor would overload with energy that can't be resolved properly. You'd end up with a completely washed out picture.

you aren't limiting resolution. You are limiting bandwidth. Bandwidth may be resolved into what is perceived as resolution but the photons are all the same size, you are just limiting photons of higher or lower frequency. As stated before, resolution is not a function of frequency it is a function of photons hitting pixels in patterns. A single photon will stimulate a pixel that is sensitive to a small range of frequency. Many photons of varying frequencies will stimulate different pixels and thus will be reconstructed as being a certain color. When you fold frequencies down into the spectrum of interest you are creating a frequency mixing situation. If you've done any work with frequencies/harmonics then you know that when you add two frequencies that are aligned in polarity, then you double their power levels. If you add two frequencies that are reverse polarity of each other, they subtract even up to the point of nulling.

Because of this you are technically getting harmonic distortion in the upper frequency range of usage. This leads to a sense of false contrast and detail. The pixels don't know the difference between the frequencies, they just get saturated and your contrast between pixels goes up.

Detail is nothing more than the power level of a reconstructed pixel. In other words, detail is really just the power level contrast between two pixels. What you are doing with a much lower cutoff frequency is getting rid of harmonics. Since the sensors are doing pixel skipping, your perceived detail is dropped considerably.

Think of the reverse of an "unsharp mask" or "sharpen" filter in your favorite NLE. Those mathmatically add contrast along edges to create the sense of detail, thus changing your perception of resolution without actually adding any real detail.

 

[Rick Burnett]

Exactly, the problem with talking about filters with light is that you need more detail as to what specific attribute of the light you are working with, as pointed out by Martti.

The IR filter deals with the light wavelength as a whole, and is a low pass filter. This I am sure makes complete sense to you.

The OLPF deals with the direction of photons through the filter. The problem is, you are thinking of light as a frequency waveform, but in this case we are looking at the direction of photons. The diffusion is spreading (blurring) the photons which prevents sharp changes in luminance (sharpness). The reason for this is that you never want the detail coming in to be finer than the sensor recording it or you get moire and aliasing in the image.

If you were to take a single line of the captured image, and graphed that, what you'd see is anything that was a square wave coming into the filter essentially will come out smoothed out more towards a sine. It's considered optical because from an optical standpoint, you've reduced the frequency of photons coming in if you treat say a horizontal or vertical slice as a waveform, which is valid.

I too have an EE degree and have taken classes on optics Although professionally I am into microelectronics, and work with CMOS designs all day!

 

[Svart]

That's because you are confounding wavelength low-pass and detail low-pass.

those are one in the same. mathmatically(sorry that's how I learned to think) higher frequencies have shorter wavelengths and therefor when you look at colors, those with lower frequencies(longer wavelengths), like red, will cover more area. Those at higher frequencies(shorter wavelengths) like blue, will cover much less area even though their power densities might be identical. This gives an object like a CMOS sensor, which has an equal density of pixels that are sensitive to red and blue, the perception that the blue has more detail, because the blue photons are covering much less area.

Our retinas are the same but we actually have more sensitivity to blue than any other color. Our rods sense achromaticly, and they are far more sensitive than the color cones. This makes the contrast between light and dark much more important to our perception of detail.

The IR filter deals with the light wavelength as a whole, and is a low pass filter. This I am sure makes complete sense to you.

Actually the IR filter is a high pass filter. the OLPF is technically a UV filter.

The diffusion is spreading (blurring) the photons

Think of pool/billiard balls. Once an object hits another object you will LOSE power in the initial object and gain power in the second. This in photons would cause frequency shifting and you would see false coloration. In an OLPF you are doing this with photons above your frequency cutoff point but any photons below that frequency are allowed to pass.

this is that you never want the detail coming in to be finer

How can you get finer detail out of an array of pixels that are all of the same size? How can you get more detail out of photons that are all the same size? You can't. detail is directly related to the contrast between pixels. Yes, blurring the visible light range would reduce detail but that's not how an OLPF works. IF it were to work like that, you'd lose power in your frequency range and your image will begin to degrade in other ways than what we are discussing. This of a ground glass in a DOF adapter. You lose light through scattering.

I too have an EE degree and have taken classes on optics Although professionally I am into microelectronics, and work with CMOS designs all day!

Awesome! I now do RF mixer designs.. So I hope I know a bit about filtering, mixing, anti-aliasing filters and the like..

 

[shaocaholica]

...stuff...

What you're talking about only applies at a very very small distances. The photosites on the sensor are not small enough. Did you read the wiki article?

http://en.wikipedia.org/wiki/Anti-aliasing_filter

The typical implementation in digital cameras is two layers of birefringent material such as lithium niobate, which spreads each optical point into a cluster of four points.[1]

It clearly describes how it works. Sure there are other sciences that apply but we're talking about the status quo implementation.

 

[dadoboy]

This was my assumption too. Isn't the whole idea of putting the OLPF from a D30 going in the wrong direction? The OLPF works on the entire 18MP of the t2i's chip, before the line skipping. The D30 OLPF would be working on the wrong frequencies/details. We'll find out soon enough.

The reason for this is that you never want the detail coming in to be finer than the sensor recording it or you get moire and aliasing in the image.

 

[Rick Burnett]

You keep treating this like a waveform issue with the OLPF does not care about the frequency of the photons AT ALL. It has nothing to do with wavelength. It has to do with direction.

I think you need to go read about moire and aliasing to understand what this filter is doing. It's just spreading out the photons to prevent sharp changes in contrast spatially. This is necessary if the density of photosites is not fine enough to capture the detail coming in, which is the case with the 7D since it line skips and has an OLPF designed for the full resolution of the sensor.

 

[Svart]

You keep treating this like a waveform issue with the OLPF does not care about the frequency of the photons AT ALL. It has nothing to do with wavelength. It has to do with direction.

I think you need to go read about moire and aliasing to understand what this filter is doing. It's just spreading out the photons to prevent sharp changes in contrast spatially. This is necessary if the density of photosites is not fine enough to capture the detail coming in, which is the case with the 7D since it line skips and has an OLPF designed for the full resolution of the sensor.

Actually I respectfully feel the same way about those who think this filter does anything but low passing the frequency range.

I think you fail to understand that frequency is exactly the attribute that is being modified. Maybe I can't explain it well enough. It's not a simple subject to try to explain in layman's terms.

 

[shaocaholica]

This was my assumption too. Isn't the whole idea of putting the OLPF from a D30 going in the wrong direction? The OLPF works on the entire 18MP of the t2i's chip, before the line skipping. The D30 OLPF would be working on the wrong frequencies/details. We'll find out soon enough.

Think of it this way. The 7D/T2i reads one line and skips ~9. With the D30 OLPF, those 10 lines will get blurred into one line so when the sensor goes to read just one line, its actually reading all 10 as a blur. Its almost the same as reading all 10 lines and then reducing all that data into one line a la super sampling.

 

[Svart]

What you're talking about only applies at a very very small distances. The photosites on the sensor are not small enough. Did you read the wiki article?

http://en.wikipedia.org/wiki/Anti-aliasing_filter



It clearly describes how it works. Sure there are other sciences that apply but we're talking about the status quo implementation.

Sure, I know all about it. I use it every day.

Did you read how it applies to optical and camera implementations?

http://en.wikipedia.org/wiki/Nyquist%E2%80%93Shannon_sampling_theorem#Aliasing

 

[shaocaholica]

Sure, I know all about it. I use it every day.

Did you read how it applies to optical and camera implementations?

http://en.wikipedia.org/wiki/Nyquist%E2%80%93Shannon_sampling_theorem#Aliasing

Guh, I'll read it over lunch. Making me activate all those brain cells left dormant after my BS. :S

 

[Rick Burnett]

Oops, yeah, I meant to say that the IR filter was a HP filter my mistake there.

But the OLPF is NOT a LP filter of frequency of light. You are very confused here. It diffuses light at all frequencies by spreading it out which essentially causes some of the light to go to adjacent photosites on the CMOS sensor. It HAS to do this without regard to frequency or it would affect the color of the image, and it does not.

Light passes through clear glass all the time and doesn't change color.

This is how a OLPF works. You are confusing signal theory with optical spacial issues, which are different. Go here:

http://en.wikipedia.org/wiki/Anti-aliasing_filter

And read the second part where the OLPF is also called a blur filter. It's doing exactly as I have stated.

As I have built a DoF adapter, the purpose of the ground glass has NOTHING to do with this. It's a focusing screen so the video camera can record the image. You've got two different lens systems and this is a common point in the center so the 35mm lens can project it's image on the focusing screen (ground glass) and the video camera lens can record that image.

The OLPF does degrade the image, that's the point! The sensor in the camera cannot record the detail coming in with the assumption the lens resolution is greater than the sensor. If you used a lens that resolved less than the resolution of the sensor, you wouldn't need a OLPF.

Again, you are mixing frequency with spacial issues.

And to give you a little more to chew on, photons are massless, when they bounce off of something, they don't change frequency, only direction.

 

[Svart]

The short of it is that the filter is not "blurring" an image as a foggy window would blur your view of the outside world. It's "blurring" by cutting off higher wavelengths. Since those wavelengths are much shorter, they are more selective in their stimulation of pixels.

Think of the difference between using a spray-paint can close up(short wavelength) and far away(long wavelength). The short wavelength will give you a much cleaner line while your longer wavelength will give you a more diffuse line over a larger area.

I'm sure you've read that the different color pixels in these sensors have different sensitivities. That's by design. That's so you can get the same output from the different color groups even though their power levels are NOT the same coming into the camera.

Add in the fact that you are mixing down additive frequencies which make your blue pixels more excited and you have false detail.

Ever wonder why the strange colors you see in moire patterns are usually blue/red and purple? Red is the range that gets mixed up from IR, blue is the range that gets mixed down from UV. Purple is the mix between the two false colors.

 

[Svart]

And to give you a little more to chew on, photons are massless, when they bounce off of something, they don't change frequency, only direction.

Tell that to the guys using blue-shifting and gravitational lensing in astronomy..

If something is modified by e-mag fields then it has mass. Just because we can't measure it doesn't mean it doesn't have mass.

I also implore you to take a look at the frequency plots in this datasheet: https://www.avxcorp.com/docs/catalogs/opticaldev.pdf


Anyway, I'm not arguing that an OLPF doesn't work by "blurring", I'm saying that it's not doing it by the way that people think. It's doing it the same thing as listening to your speakers with cotton stuffed in your ears. You lose the details in the audio because your higher frequencies are being cut.

 

[shaocaholica]

I also implore you to take a look at the frequency plots in this datasheet: https://www.avxcorp.com/docs/catalogs/opticaldev.pdf

You should look closely at that too. Look at the figure showing the OLPF stack. There are 2 refractive layers and a IR cut layer. The refractive layers are doing the "blurring". The frequency graphs show how the IR cut layer works but its not the only thing in that stack.

Anyway, I'm not arguing that an OLPF doesn't work by "blurring", I'm saying that it's not doing it by the way that people think. It's doing it the same thing as listening to your speakers with cotton stuffed in your ears. You lose the details in the audio because your higher frequencies are being cut.

But all these sources are saying the "blurring" effect is due to diffraction, not frequency attenuation.

 

[Svart]

Uh.. it's exactly as I said. The UV LPF are done by refraction. The IR cut is done by a reflective coating. Not sure what you are trying to say.

But all these sources are saying the "blurring" effect is due to diffraction, not frequency attenuation.

Because they are speaking in generalities, not specifics.

Look at the graphs in the datasheet. You'll see that at below 400nm you have a very sharp cutoff on the OLPF graph. On the IR only graph you don't have that cutoff. However the optical passband transmittance % is exactly the same.

IF the OLPF scattered visible light to "blur" as you try to explain it, your passband transmittance would drop significantly and there would be no UV cutoff.

 

[Rick Burnett]

But you are talking about two different types of frequencies, they are NOT the same.

One frequency is of the photons coming in. This is the color that the photon would represent (or in the case of IR that we cannot see). The IR filter reflects back photons below (thanks for correcting me) a certain photon frequency, anything in the IR range or lower.

The OLPF does not care AT ALL about the frequency of the photon. You are absolutely incorrect about this. The issue with harmonics and getting their (in terms of audio) beat frequencies and such in the visible range are not what the OLPF is there for, at all. You need to go read the section on "Application to multivariable signals and images again" and see that it talks about "sampling in the spacial domain" and "images can also suffer from aliasing if the sampling resolution, or pixel density, is inadequate" This has NOTHING to do with the waveform frequency of a photon AT ALL. And the solution, in the article you referenced is "or to optically blur the image before acquiring it with the sensor."

Pertaining to UV light, each pixel has a filter on it that is a band pass essentially, so this is not needed.

You are mixing a whole bunch of electron theory with photon theory. An active pixel sensor measures photons hitting it, like a bucket. It could care less about frequency. That's why a R, G or B filter is before them so only wavelengths in those ranges come through. I imagine the purpose for the IR filter is that the R G B filters do not filter IR themselves or it wouldn't be needed either.

If the sensor passes UV information though any of the pixels, it will be added to the bin. It's not going to create harmonics because this is not the physics that are going on here. The frequency does not matter to the active pixel. It counts photons regardless of frequency. Period.

To put this in other terms, sampling in video refers to more than one aspect of sampling. There is a color segment and a spatial segment. To make an analogy to the audio world, let's assume you have to microphones that only detect sound in one direction. You have a speaker between them that shoots audio exactly in a straight line. We are in an anechoic chamber if you will.

If you send out a sound, nothing is recorded on the two microphones because they are not in the path of the sound.

Now, you place a device in front of the speaker that cause the sound to reflect towards the microphones. Yes, you have reduced the sound level since you have split it into two sides, but now you get that sound recorded and represented in the recording. Did you change the frequency of the sound? No

This is a VERY GENERIC example of what we are trying to accomplish. We are overcoming a spacial sampling limitation of the sensor that has NOTHING to do with the frequency of the waveforms AT ALL.