Welcome to our community

Be a part of something great, join today!

Log in
  • Hey all, just changed over the backend after 15 years I figured time to give it a bit of an update, its probably gonna be a bit weird for most of you and i am sure there is a few bugs to work out but it should kinda work the same as before... hopefully :)

Is this chroma aliasing?

Patrick Tresch

Patrick Tresch

Well-known member
Joined
Aug 2, 2007
Messages
5,808
Reaction score
86
Points
48
Age
55
Location
Switzerland, Lausanne
Hello,

I've seen some strange chorma aliasing, I've never seen before on a RED camera (I think I would have paid more attention to it).
Is this due to some Olpf/highIso/FW combination?



See you.

Pat

Shot on Celere HS 18,5mm lens @ 1600iso

https://wetransfer.com/downloads/60...7e4973b34b2bafeb250f765f20170428095730/7a9b48

CelereHs_chart.zip are with R3D
Compression_Test.zip are some jpeg compression tests from 4:1 to 22:1
 
Patrick Tresch said:
Hello,

I've seen some strange chorma aliasing, I've never seen before on a RED camera (I think I would have paid more attention to it).
Is this due to some Olpf/highIso/FW combination?



See you.

Pat

Shot on Celere HS 18,5mm lens @ 1600iso

https://wetransfer.com/downloads/60...7e4973b34b2bafeb250f765f20170428095730/7a9b48

CelereHs_chart.zip are with R3D
Compression_Test.zip are some jpeg compression tests from 4:1 to 22:1
Click to expand...

Whoa...that's an terribly looking 1024x626 pix JPEG, made probably out of 3840x2346pix JPEG. BTW, You're at 200%...and in RCXP 42! :001_smile:

Here, an 1920x1080 pix, 1:1 (100%) TIFF out of RCXP 50 - beta1:

https://drive.google.com/file/d/0B_Fxz1xiV9sjcmNCNG42RTllZTQ/view?usp=sharing

No aliasing...:biggrin:
 
Patrick Tresch said:
Hello,

I've seen some strange chorma aliasing, I've never seen before on a RED camera (I think I would have paid more attention to it).
Is this due to some Olpf/highIso/FW combination?




See you.

Pat

Shot on Celere HS 18,5mm lens @ 1600iso

https://wetransfer.com/downloads/60...7e4973b34b2bafeb250f765f20170428095730/7a9b48

CelereHs_chart.zip are with R3D
Compression_Test.zip are some jpeg compression tests from 4:1 to 22:1
Click to expand...

Those Rokinons, emmm....:devil::devil: Celeres may have some color to them.
Just kidding- I don't know for sure where the glass comes from. This is pure conjecture to be clear.
They could be more transparent and less defensive about it- that attitude does not help. Internet and Google images are great equalizers these days...
Can't hide much any more...
Anyway....Some lenses wide open may exhibit quite a bit of lateral dispersion that may look like color moire in some cases....
 
It sure looks like it.

Red claims it's getting nearly 80% luma resolution from a bayer sensor or something close to it and that just defies the nyquist sampling theorem if it's without aliasing.

I see a lot of aliasing (including luma aliasing in red fabrics) with the Alexa and C300, the cameras I work with most. I've seen less with the Red but use it less, and it is softer per-pixel (but with more pixels). That's not so bad I the scheme of things. Chroma aliasing is easier to fix in post and this is a pretty extreme situation that will create it.

That said if you're processing or viewing at all wrong the aliasing could be from that.
 
"Red claims it's getting nearly 80% luma resolution from a bayer sensor or something close to it and that just defies the nyquist sampling theorem if it's without aliasing. " - we never defy Nyquist.

Graeme
 
Graeme Nattress said:
"Red claims it's getting nearly 80% luma resolution from a bayer sensor or something close to it and that just defies the nyquist sampling theorem if it's without aliasing. " - we never defy Nyquist.

Graeme
Click to expand...

Lol. I mean, no one does. Because you can't. I'm just saying you have to make a trade off and if you're resolving more than 50% of resolution (with bayer even less since pure red or blue resolution is 50% of that) you're gonna get some aliasing somewhere. The Alexa has a lot of aliasing, but it rarely bothers me. Red seems to have even less but that doesn't mean it's not there in extreme situations like test charts. :)

Unless I'm mistaken... I don't think I am. But I'm not the expert on this either.
 
Last edited:
Red cameras have the lowest aliasing in their class (and beyond). But that doesn't mean they have none. You can provoke color aliasing with critical test patterns. If they'd filter for no chroma aliasing on a Bayer sensor, everybody would blame them for being too soft. Even now some Sony or BlackMagic users say that Red cameras are soft. But, hey, look at their aliasing! F3 in particular and all BM cameras. To completely filter out chroma aliasing you'd need 4 times the resolution, which would cost too much in processing power and sensitivity.

I've never seen luma aliasing with a Red (and have been around since R1) and very rarely small traces of chroma aliasing in regular shooting. But every camera has to keep a balance in this and many other respects and IMHO Red is doing pretty well there.

BTW, to provoke aliasing on a Red you'll need some razor sharp glass!
 
Matt W. said:
Lol. I mean, no one does. Because you can't. I'm just saying you have to make a trade off and if you're resolving more than 50% of resolution (with bayer even less since pure red or blue resolution is 50% of that) you're gonna get some aliasing somewhere. The Alexa has a lot of aliasing, but it rarely bothers me. Red seems to have even less but that doesn't mean it's not there in extreme situations like test charts. :)

Unless I'm mistaken... I don't think I am. But I'm not the expert on this either.
Click to expand...

Let's start with the simplest case - a monochrome sensor. Say it's a 4k sensor. Sampling theory would say you need a sampling frequency at least double that of the sampled waveform to avoid aliasing. In other words, you need at least two samples per wavelength. If we think of a wavelength as a pair of lines, one white, one black we can re-state it as saying we need at least two samples per line pair. As a line pair contains two lines, we can now see that we need at least two samples per two lines, or a sample per line. A 4k mono sensor has 4k lines and thus there is no violation of sampling theory by getting 4k resolution from it (strictly < 4k resolution would be correct, but any resolution up to but not including 4k would be valid).

As you can see the confusion comes in through sampling theorem discussing frequencies (and hence entire wave lengths) compared to samples by a factor of 2, and a wavelength in the imaging sense is two lines.

Bayer gets more complex. For red and blue you have half the resolution before aliasing as compared to a mono sensor. But good image processing can minimize the appearance of that, as we do.

For green which relates to what we normally think of as "resolution" in an image, we have 1/2 the samples of the mono sensor (in an area sense) or 1/root(2) in a linear sense, so a simple approach would give you a max rez before aliasing of around 74%. However, that's where demosaic algorithms come it and can use the red and blue data to vastly increase the effective sample points.

Of course, we don't aim (even on a mono sensor) for 100% of the possible resolution because we need the low pass filter. But unlike what you can do with software and electronic filters, optical low pass filtering is "slow" and we have to cut early to get significant attenuation where we need it.

So no, we don't violate sampling theory on the luma resolution side of things, and on chroma, like everyone else using bayer patterns, we make an educated compromize between aliasing and resolution.

Graeme
 
A question for Graeme? Would it be fair to say; HRR NOTE: 100% Advertised Resolution Less 74% = Actual Resolution?

Humberto Rivera
 
Love to read stuff, that makes no sense at all :-)

Graeme Nattress said:
Let's start with the simplest case - a monochrome sensor. Say it's a 4k sensor. Sampling theory would say you need a sampling frequency at least double that of the sampled waveform to avoid aliasing. In other words, you need at least two samples per wavelength. If we think of a wavelength as a pair of lines, one white, one black we can re-state it as saying we need at least two samples per line pair. As a line pair contains two lines, we can now see that we need at least two samples per two lines, or a sample per line. A 4k mono sensor has 4k lines and thus there is no violation of sampling theory by getting 4k resolution from it (strictly < 4k resolution would be correct, but any resolution up to but not including 4k would be valid).

As you can see the confusion comes in through sampling theorem discussing frequencies (and hence entire wave lengths) compared to samples by a factor of 2, and a wavelength in the imaging sense is two lines.

Bayer gets more complex. For red and blue you have half the resolution before aliasing as compared to a mono sensor. But good image processing can minimize the appearance of that, as we do.

For green which relates to what we normally think of as "resolution" in an image, we have 1/2 the samples of the mono sensor (in an area sense) or 1/root(2) in a linear sense, so a simple approach would give you a max rez before aliasing of around 74%. However, that's where demosaic algorithms come it and can use the red and blue data to vastly increase the effective sample points.

Of course, we don't aim (even on a mono sensor) for 100% of the possible resolution because we need the low pass filter. But unlike what you can do with software and electronic filters, optical low pass filtering is "slow" and we have to cut early to get significant attenuation where we need it.

So no, we don't violate sampling theory on the luma resolution side of things, and on chroma, like everyone else using bayer patterns, we make an educated compromize between aliasing and resolution.

Graeme
Click to expand...
 
Humberto Rivera said:
A question for Graeme? Would it be fair to say; HRR NOTE: 100% Advertised Resolution Less 74% = Actual Resolution?

Humberto Rivera
Click to expand...

We advertise the resolution of the sensor - as does everyone else. Measured resolution is something we talk about if asked and for them we shoot sine zone plates and measure MTF, and quote the 10% MTF value for limiting resolution.

Graeme
 
So possibly; 8,000,000 X 75% = 6,000,000 or 6K of Actual Resolution or thereabouts?

Humberto Rivera
 
Humberto Rivera said:
So possibly; 8,000,000 X 75% = 6,000,000 or 6K of Actual Resolution or thereabouts?

Humberto Rivera
Click to expand...

The green channel itself is exactly 6k, so to follow what he was saying by using red and blue you should be able to get a little more.

Of course "resolution" doesnt make sense without a MTF reference point. If a black and white line pair are reproduced as 80% grey and 20% grey pairs, are they resolved?
 
Graeme Nattress said:
Let's start with the simplest case - a monochrome sensor. Say it's a 4k sensor. Sampling theory would say you need a sampling frequency at least double that of the sampled waveform to avoid aliasing. In other words, you need at least two samples per wavelength. If we think of a wavelength as a pair of lines, one white, one black we can re-state it as saying we need at least two samples per line pair. As a line pair contains two lines, we can now see that we need at least two samples per two lines, or a sample per line. A 4k mono sensor has 4k lines and thus there is no violation of sampling theory by getting 4k resolution from it (strictly < 4k resolution would be correct, but any resolution up to but not including 4k would be valid).

As you can see the confusion comes in through sampling theorem discussing frequencies (and hence entire wave lengths) compared to samples by a factor of 2, and a wavelength in the imaging sense is two lines.

Bayer gets more complex. For red and blue you have half the resolution before aliasing as compared to a mono sensor. But good image processing can minimize the appearance of that, as we do.

For green which relates to what we normally think of as "resolution" in an image, we have 1/2 the samples of the mono sensor (in an area sense) or 1/root(2) in a linear sense, so a simple approach would give you a max rez before aliasing of around 74%. However, that's where demosaic algorithms come it and can use the red and blue data to vastly increase the effective sample points.

Of course, we don't aim (even on a mono sensor) for 100% of the possible resolution because we need the low pass filter. But unlike what you can do with software and electronic filters, optical low pass filtering is "slow" and we have to cut early to get significant attenuation where we need it.

So no, we don't violate sampling theory on the luma resolution side of things, and on chroma, like everyone else using bayer patterns, we make an educated compromize between aliasing and resolution.

Graeme
Click to expand...

I love when you get technical :)
 
Graeme Nattress said:
Let's start with the simplest case - a monochrome sensor. Say it's a 4k sensor. Sampling theory would say you need a sampling frequency at least double that of the sampled waveform to avoid aliasing. In other words, you need at least two samples per wavelength. If we think of a wavelength as a pair of lines, one white, one black we can re-state it as saying we need at least two samples per line pair. As a line pair contains two lines, we can now see that we need at least two samples per two lines, or a sample per line. A 4k mono sensor has 4k lines and thus there is no violation of sampling theory by getting 4k resolution from it (strictly < 4k resolution would be correct, but any resolution up to but not including 4k would be valid).
Click to expand...

That actually makes some sense, but what I fail to understand is this: let's say we're in a perfect world with a perfect system between the sensor and the subject, everything else in the chain is 100% mtf exactly, and we have a 4k sensor grid and a 4k printed image of 2048 line pairs. Let's line them up perfectly. Then we have up to 100% mtf of 2048 line pairs. The captured image is the same as the printed one.

Now let's move the camera a bit so that the sensor is offset by just one half pixel. Now, we're suddenly seeing gray at every photo site. And we have 0% mtf.

Are you implying that by removing just one line pair from that image, this phenomenon will no longer occur in this way? It seems like it still would but the pattern would be wavy rather than being straight 100% mtf line pairs transitioning to 0% mtf (pure gray) across the frame no matter how exactly it's placed. It seems to me like you'd have to remove half the line pairs from that image so that the frequency of the image itself was low enough that this phenomenon didn't occur at all and you could approach 100%mtf capture at any frequency. How is this not the case? Or is aliasing simply when you get erroneous results of >100% mtf?
 
Last edited:
You must log in or register to reply here.
Back
Top