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

[Svart]

Frequency is frequency, I don't see how you can discern that there is more than one kind..

I still think you are missing what I am saying though. Maybe I'll think about another way to explain it.

I do understand what you are saying but I think you are looking at it wrong. think about it in the frequency domain, not time domain.

Anyway, I'm out of time for discussion today I have to get to work.

 

[Rick Burnett]

One question:

Do you understand the difference between spatial frequency and light (wavelength) frequency?

Update: Yes, the actual concept of frequency is the same, but there are multiple characteristics being measured in an optical system and you are mixing them together like they are the same thing. They are in fact absolutely different physical properties/characteristics that have cyclical properties being measured. Frequency in and of itself does not dictate that the cause for the property can be modified the same way. That's absurd.

Once you understand the difference between what spatial frequency in an optical system represents and light frequency, you realize they are two different characteristics of optics that are dealt with separately for different reasons in an optical system. I can think in the frequency or time domain all day long, it doesn't change the different between an IR filter and a OLPF filter.

 

[Svart]

OK one more answer:

Yes.

"Spacial" frequency: period of a repeating object
"wavelength" frequency: Spacial period of a sine waveform
"temporal" Frequency: number of repeating events in a given time

Those are identical definitions. It is a measure of a repetitious action vs. time.

It could be how many times a photon vibrates vs. time(where time is the control) or it could be how many pixels are in an array(where time is understood to be 0 )

either way, its still the same definition. "time" is the only attribute that differentiates the two.

I understand what you are trying to say but I think you are not looking at the theory/math close enough.

 

[Rick Burnett]

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.

This makes absolutely no sense at all.

 

[Svart]

ok one more..

How does that not make sense? They scatter the UV and reflect the IR which creates a BPF in the visible light range. How does that not make sense?

 

[Rick Burnett]

OK one more answer:

Yes.

"Spacial" frequency: period of a repeating object
"wavelength" frequency: Spacial period of a sine waveform
"temporal" Frequency: number of repeating events in a given time

Those are identical definitions. It is a measure of a repetitious action vs. time.

It could be how many times a photon vibrates vs. time(where time is the control) or it could be how many pixels are in an array(where time is understood to be 0 )

either way, its still the same definition. "time" is the only attribute that differentiates the two.

I understand what you are trying to say but I think you are not looking at the theory/math close enough.

No, actually, you don't have a clue what you are talking about. You are posting generic terminology that isn't rooted in optical theory for which the applied concept of frequency is being used to represent.

Let me help you.

There are three different properties of frequency with an optical system that is being sampled by pixels.

(1) Color of the pixel. This is a wavelength/frequency. (multiple ways of describing).

(2) Change of color in a single instance of time across pixels in a straight line. If you took a row of pixels, you could plot this as a waveform. This is spacial frequency. It deals with adjacent pixels.

(3) Change of color of a single pixel with respect to time.

These three characteristics, which can be described by frequency, affect an optical system in totally different ways. They represent different characteristics and each while pulled from some of the same data, have additional data that are the reason they are different.

(1) is the purpose for an IR filter or a UV filter. It is also the purpose for the R,G and B filters over the active pixel site. Pixel sites measure photons, not frequency. They are filtered so each pixel site picks up a primary color so the image can be represented in a system that uses the same technology to reproduce color (RGB) or can at least be converted to a method that is compatible. No color information is recorded directly by a CMOS sensor, you have to know which optical filter was above your active site to know what color it represents.

(2) is the purpose for a OLPF, or in layman terms, a filter that blurs an image regardless of frequency. The spacial resolution of pixels sites is not fine enough to capture the contrast changes in an image in space (X,Y for simplicity) so the image is blurred to make sure fine detail is not lost but spread out into the active pixel area to avoid moire/aliasing.

(3) is the frequency change that pixels are changing per second. This is the most analogous to audio capture. The perfect example of this is try recording a light flickers at faster than your frame rate just slightly. Sometimes the light will be on, sometimes off, but it will not be constant like what you see in real life. We don't filter for this in optical systems, but it's the closest thing to harmonic issues like you describe that you see in audio systems.

It's that straightforward. If you still cannot get these concepts, I give up.

 

[Rick Burnett]

ok one more..

How does that not make sense? They scatter the UV and reflect the IR which creates a BPF in the visible light range. How does that not make sense?

Where do they scatter to?

 

[Svart]

While I'm supposed to be working since my lunchtime is over, I'll make this brief.

Yes I understand what you are talking about but I still don't think you are even attempting to understand what I am saying.

Pixel sites measure photons, not frequency.

But the frequency determines the color. You said it yourself. If a pixel didn't care what "color" photon stimulated it, then everything would be achromatic. The pixels DO care what the frequency is. Some sensors dope the junctions to achieve frequency range sensitivity, some use physical filtering, etc. ALL of them use external physical filtering to lessen the power in the sensitive band and reject aliased images. There is no argument about this. This is the reason for the OLPF, no amount of obfuscation can change the fact that you are aliasing in the physical domain and these filters stop that

Admittedly I don't know what type of CMOS sensor that the Canon cameras use but there are a few different kinds that work in different ways. If you could specify which type of sensor that canon uses, we could discuss if in more detail at a later time.

Where do they scatter to?

It depends on the refraction index of the quartz, which is almost assuredly proprietary. Either way you are scattering light as you would with any translucent material. The angle of incidence and the refraction index would determine how the light is scattered. Effectively you are "bouncing" the light around in the material until it loses it's energy(becomes a lower frequency).

 

[Rick Burnett]

] Some sensors dope the junctions to achieve frequency range sensitivity, some use physical filtering, etc.

Please feel free to point out a high quality CMOS sensor that does this. Or better yet, please show me where doping is changed in the photodiode in practice.

There are patents for generating color detection through multiple photodiodes by varying the back bias of the substrate differently for each of the photodiodes. This changes the depletion area of the diode electrically. The whole point of using CMOS is not to use special fabrication techniques which is the advantage over CCD design.

EVERY high end CMOS sensor I have seen or researched includes a color filter layer above the CMOS sensor. I've seen no supporting evidence that any other method of CMOS design produces a better image or color separation. And by color separation I mean the filter AND the CMOS sensor used in conjunction.

But regardless, who cares! We are talking about the 7D/550D/t2i at the moment which we know uses color filters and a bayer pattern.

The OLPF does not scatter UV light, it blurs ALL light coming through it. Which is why using a UV filter on the front of your lens does affect the image.

 

[Rick Burnett]

It depends on the refraction index of the quartz, which is almost assuredly proprietary. Either way you are scattering light as you would with any translucent material. The angle of incidence and the refraction index would determine how the light is scattered. Effectively you are "bouncing" the light around in the material until it loses it's energy(becomes a lower frequency).

Honestly, I'm done trying to explain this to you. You don't get it. That's not what is happening nor what the purpose of an OLPF is for. It has nothing to do with UV light AT ALL. Nor has it anything to do with photon frequency AT ALL. It's a correction for spacial sampling limitations in sensor pixel densities.

 

[ydgmdlu]

Svart is wrong, and even a layman (like me) can prove it quite easily.

Optics company Sunex has this explanation of optical low-pass filters:

In high-quality digital imaging systems, optical low-pass filters (OLPF) are used to eliminate color Moire fringes. An OLPF cuts off the lens MTF above the sampling frequency of the imager resulting an overall MTF curve that approximates a step function in spatial domain.

So the key here is MTF (modulation transfer function). Here's what Wikipedia says about MTF:

The optical transfer function (OTF) describes the spatial (angular) variation as a function of spatial (angular) frequency. When the image is projected onto a flat plane, such as photographic film or a solid state detector, spatial frequency is the preferred domain, but when the image is referred to the lens alone, angular frequency is preferred.
...
The OTF accounts for aberration. Its magnitude is known as the Modulation Transfer Function (MTF) and its phase is known as the Phase Transfer Function (PTF).

So now we have to see what "spatial (angular) frequency" means. From the Wikipedia article:

In mathematics, physics, and engineering, spatial frequency is a characteristic of any structure that is periodic across position in space. The spatial frequency is a measure of how often the structure repeats per unit of distance. The SI unit of spatial frequency is cycles per meter. In image processing applications, the spatial frequency often is measured as lines per millimeter, which is 1000 times smaller than the SI unit.

Based on the above, the following comments would seem to be wrong:

"Spacial" frequency: period of a repeating object
"wavelength" frequency: Spacial period of a sine waveform
"temporal" Frequency: number of repeating events in a given time

Those are identical definitions. It is a measure of a repetitious action vs. time.

It could be how many times a photon vibrates vs. time(where time is the control) or it could be how many pixels are in an array(where time is understood to be 0 )

either way, its still the same definition. "time" is the only attribute that differentiates the two.

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.

Since wavelength determines color, what you are saying here is that an OLPF is actually a color filter, which is patently ridiculous.

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.

The aliasing in the video image is caused exclusively by the camera's downscaling algorithm. If what you were saying were true, then there would egregious aliasing all over every full-resolution still image from the camera. But that's not the case. It only appears in video, due to the downscaling (possibly line-skipping) that the camera does.

 

[Martti Ekstrand]

Frequency is frequency, I don't see how you can discern that there is more than one kind..

Well, it's the tiny little matter of what is measured in frequencies. Are you saying that the frequency of how often the garbage guys come by and pick up my trash is the same kind as the frequency of how telephone poles are placed in my street? In other words temporal vs spatial.*

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.

Funny, this is exactly what it does.

The 'frequency' that this filter deals with is the distance between details in the 2D image projected on the sensor plane versus the distance of the photo diodes, ie; it should mush together the fine details than the ends up between the photo diodes to minimize aliasing. So again: you confound wavelength temporal frequency with detail spatial frequency.

The hard trick here is to make a filter that blurs just the same amount as the distance between the photo diodes. Blur too much and you have a image on the run from a 70s shampoo commercial, blur too little and it doesn't do much at all to fix aliasing.

*There also this, http://en.wikipedia.org/wiki/Urinary_frequency but since obviously you are a young man leave that to us ol' gaffers.

 

[Rick Burnett]

The aliasing in the video image is caused exclusively by the camera's downscaling algorithm. If what you were saying were true, then there would egregious aliasing all over every full-resolution still image from the camera. But that's not the case. It only appears in video, due to the downscaling (possibly line-skipping) that the camera does.

First, thanks for your response, I was on those pages too looking at that particular OLPF but was running out of steam trying to explain these things.

The reason it does not appear in stills is because the OLPF on the sensor is designed for the full resolution of the sensor. The reason for this whole experiment in the first place was to instead replace the OLPF with one that would be for 1080p which is our desired capture rate. However, the actual resolving power of the sensor seems to be lower than 1080p which some have questioned if even a stronger OLPF should be used since test charts have shown the actual detail resolution of the 7D to be closer to 720p.

In addition, if this LP filter works for the 1080p mode, it will surely soften the image, and will reduce the problem in 720p mode, it will not eliminate them. Canon's downsampling is just skip lines and ignore other pixels it appears.

The ultimate question is will that softening of the image still be pleasing or is it the false detail that is the ultimate plus and minus to the 7D

(And yes, line skipping is the culprit)

 

[Barry_Green]

Grimepoch has the purposes down exactly right. Yes, the idea is to see whether the t2i becomes an alias-free (or much-more-alias-free) video camera. Shaocaholica's losing the ability to get high-resolution stills, but the experiement here is to see if we could find a more appropriately-matched OLPF for video-only purposes.

 

[Svart]

An OLPF cuts off the lens MTF above the sampling frequency of the imager resulting an overall MTF curve that approximates a step function in spatial domain.

I don't know what you guys are smoking but re-read what I said. This quote is exactly what I explained. Frequencies above the sampling frequency(those above the useful bandwidth, I.E. UV) are aliased down into the visible spectrum. According to the MTF page on wiki, when you plot the FFT of this transfer on a log-frequency plot(Bode) you get the resulting LOW FREAKING PASS filter knee. How many times do I have to say it? It's a LOW PASS FILTER. BY FILTERING HIGH FREQUENCIES you are effectively "blurring" your edges.

From the same MTF wiki page:

The modulation transfer function represents the Bode plot of an imaging system (such as a microscope or the human eye), and thus depicts the filtering characteristic of the imaging system. The human eye, for instance, acts as a low-pass filter, meaning that very high-frequency components (sharp edges) cannot be perfectly perceived.

Are you saying that the frequency of how often the garbage guys come by and pick up my trash is the same kind as the frequency of how telephone poles are placed in my street? In other words temporal vs spatial.*

I learned frequency as an algebraic function. it roughly translates to words as: Period VS. Time. If you want to correlate the garbage pickup its; how many times it gets picked up VS. the length of time you want to measure, I.E., 4 times per month. If you want to look at telephone poles; how many poles VS. time, which in this case is ZERO because you see them all at the same time, I.E., 10 poles right now. The freaking math is the same, just the numbers you plug in are different.

Again:

In high-quality digital imaging systems, optical low-pass filters (OLPF) are used to eliminate color Moire fringes. An OLPF cuts off the lens MTF above the sampling frequency of the imager resulting an overall MTF curve that approximates a step function in spatial domain.

Now look at this page: http://www.brightcrystals.com/english/readnew.asp?NewsID=4903 and look at the graph at the bottom of the page. The page uses the same wording you used but this one has a graph. Look at trace C, that's the unmodified spectra. Now look at A and B. See how those traces trend downwards towards the UV range? Thats because its a LOW PASS FILTER.

Since wavelength determines color, what you are saying here is that an OLPF is actually a color filter, which is patently ridiculous.

If it cuts off frequency above the image sampling rate, then it IS cutting color. It's cutting UV, which while we can't see it, it IS a color. Birds can see in UV just as we see in "visible" light.

Your quote:

due to the downscaling (possibly line-skipping) that the camera does.

My quote:

Since the sensors are doing pixel skipping, your perceived detail is dropped considerably.

See? you are all running to condemn me yet I am saying exactly what you are saying.


So what if I can't explain things like you want to read them? I work with formulas, not words. So piss all over me if you must, but we are looking at two sides of the same coin. Physical world and electrical world are both defined by physics which is defined by math. the math is the same.

We are simply arguing over semantics at this point.


READ AND LEARN:

http://www.microsoft.com/china/whdc/archive/ovrsamp.mspx

 

[boulder]

Way to ruin an exciting thread guys

 

[shaocaholica]

Dont worry, I'll start a new thread tomorrow.

 

[Ian-T]

Ha...I wanted to say earlier "you guys GET A ROOM." But it got interesting...so I grabbed me some popcorn.

 

[meauounji]

Dont worry, I'll start a new thread tomorrow.

Thank you. As interesting as the technical side of this may be, some of us just wants to know whether it works or not.

 

[kgimedia]

Way to ruin an exciting thread guys

When I read this I wondered.

1. Is he saying the "mathy" talk is taking away from the mod/hack talk"?

or

2. The fact that they may have figured out they were talking about the same math with different words and now we have to go back to waiting for the reports on if it worked or not?

Either way I loved the discussion as it was not about which mic, memory card or lens was the best. When I learned this in my Video Engineering class, in '97 they, said CMOS was first but CCD is newer and better, so we are only gonna teach that.

As a side note I asked a current student of the same school what burst amplitude and pedestal were and he said "We are all digital now, they don't teach that stuff."