Moire Issues

[David Fairbanks]

Has anyone heard if Red has mentioned moire. It's been a issue on electronic cameras since day one as far as I know. I'm sure they will have the absolute best processing to eliminate it.

Having shot with digital stills for years, I know that the higher the resolution of the sensor, the more susceptible you are to moire in fabrics like suit jackets and the like. Smaller images from large sensors can give you moire when the actual full size image doesn't have a problem. It's even worse when the small image doesn't show anything and the full size image is covered in it.

In stills the best way to elevate it is to change focal length slightly and reshoot the shot. This wont be practical in most situations with Red.

[Brook Willard]

Hmm... perhaps there will be an optical low-pass filter over the sensor? I suppose there would be downsides...

[David Fairbanks]

Low pass... I don't see that would do anything. But I also haven't messed with low pass filters (or high pass for that matter).

[Finner]

I don't understand moire. Is that a different word for alising?

[David Fairbanks]

Moire is interference patterns.
The texture in cloth for instance will render in odd patterns when it is captured by a patterened sensor.
http://photo.net/bboard-uploads/004WFK-11367184.jpg

[Brook Willard]

An OLPF effectively blurs the image just a TINY little bit to avoid moiré. Again, it's good and bad.

[David Fairbanks]

If it's over the sensor, we wouldn't have the option of removing it. But if it's not a noticable amount of softening, it probably wouldn't matter.

Surely my D2X doesn't have it, it loves to moire.

[Brook Willard]

It's about 1-2 pixels of blur generally... I guess we'll find out.

[Anders Holck]

Every Bayer based sensor should have a OLPF that matches the sensors resolution.
Current pro DSLR's seem to have a filter that is tuned for a sharper image and will still show a little moiré in worst case situations.
The RED camera will have one as well. I believe some of the first shots previewed was without any filter, but the recent ones were with it fitted.

[Brook Willard]

Was that ever mentioned? I don't recall reading that anywhere.

[David Fairbanks]

I know some of the first images were very raw, that might include no LPf.

[Johan Lindgren]

http://www.dvxuser.com/V6/showpost.php?p=755006&postcount=33

Jarred saidd they were testing with IR and lo pass filters back in the dvxuser forum

Johlan

[Brook Willard]

Good find.

[Anders Holck]

BTW. here is a great example of Moiré vs Low pass filter
http://www.dpreview.com/reviews/kodakslrc/page22.asp
The Kodak on the left clearly resolves more detail (note that it also has a 2.5 MP advantage) but that moiré is pretty annoying.

[createra]

There is some issues with the hair in the Bubblegirl clip. But it's also mentioned that it's shot without a low pass filter.

[Stephen Williams]

Hi,

If you can turn off any sharpening (Red does not have any) you should not have big issues.

Remember the project is 'work in progress', I have not seen any problems with images so far.

Stephen

[Nick Shaw]

BTW. here is a great example of Moiré vs Low pass filter
http://www.dpreview.com/reviews/kodakslrc/page22.asp
The Kodak on the left clearly resolves more detail (note that it also has a 2.5 MP advantage) but that moiré is pretty annoying.

Is the colour artifacting you can see on the Kodak images moiré or bad de-Bayering artifacts? Moiré should not affect the colour, should it?

Nick

[David Fairbanks]

My experience is that it has nothing to do with sharpening. It's just the interference of the two patterns.

The images they have produced have been amazing, and it is a work in progress, so time will tell.

[Graeme Nattress]

Moire is visible aliassing. Being a sampled system, the sensor will exhibit aliassing on images if it is asked to record detail smaller than 1 pixel in dimension.

The above, is applicable to all sensors, including 3 chip systems, RGB silicon filter sensors etc. That's why all sampled systems need an anti-alias filter, which in a camera is called an OLPF or optical low-pass filter. The name tells you what it does. It passes though, or leave un-effected low frequencies, and attenuates high frequencies. We want to make sure than the sensor isn't asked to sample any frequencies or details so high that they would introduce aliassing.

With a bayer pattern CFA, this is even more important because moire on a bayer pattern sensor shows up as colour moire as well as luma moire. Colour moire, as shown above is quite nasty.

Now when we did our first mysterium shots, our OLPF had not arrived, so we just shot things as it, and I used a demosaic that was designed to be gentle on the eyes and remove some of the moire that could occur. It can't get everything, but was a good solution.

Now we have an OLPF in place, and we can be more free with our choice of demosaic. We don't get nasty colour moire on edges, and the slight softening makes for a very smooth, continous image, with no visible pixels or aliassing.

Aliassing makes it harder to suspend disbelief as we don't see aliassing in the real world. (we do see moire when two fine patterns are overlayed though). We don't see fine detail having steppy edges or being made of discrete pixels.

Excessive edge sharpness is know to cause judder in slow fps recording - think 24p.

Because in motion, aliassing changes on each frame, it's more noticible than on a still as it's almost as if the pixels are waving at you saying "look at me". On a still image, if there's a bit of moire, you can paint it out. On a movie, that would be terrible!

Graeme

[GlennChan]

Aliassing makes it harder to suspend disbelief as we don't see aliassing in the real world.
We do see aliasing on stuff that moves (although motion blur tends to act as a low pass filter). But other than that, for stationary stuff, we see incredibly low/no aliasing.

We also seem to get less aliasing in dim conditions. You can test this out by printing out a zone plate test pattern and holding it in your hands and moving it around. You can get one off here: (scroll down)
http://www.ecinemasys.com/products/edp100/edp100_lcd_vs_crt.htm

Or
http://www.worldserver.com/turk/opensource/#ZonePlate

[Graeme Nattress]

"We do see aliasing on stuff that moves " Can't say I've seen that myself - care to elucidate? Is that spatial (which is what I'm on about) or temporal aliassing?

Graeme

[jbeale]

You can get temporal aliasing (rotating objects seem to stop or move backwards, etc.) if you have a fluorescent light, strobe light or any other light that turns on and off, but I don't understand how spatial aliasing (moire patterns) happens with your naked eye unless you are looking at a scene through a screen door, grille, cloth etc.

In theory, since your eye's retina is composed of discrete photosensitive sites it could have aliasing, but the rods and cones have no fixed pattern, and besides your eye is in nearly constant motion. I'm curious in what situation you could possibly see spatial aliasing.

[Graeme Nattress]

And the brain has fantastic image algorithms!

Graeme

[Jared VanLeuven]

And the brain has fantastic image algorithms!

Graeme

This came to my mind the other day as I was driving along and noticed the wheels of a car in the lane next to me. Its wheels (spokes) looked like they were moving backwards, as in the same effect that you can notice on film. I wondered what our eyes' refresh rate is?

Oh, and no, the wheels in question weren't those funky spinners... :)

[Graeme Nattress]

Yes, that's temporal aliassing.

The brain is funny too though, with optical illusions and the like!

Graeme

[Jeff Kilgroe]

I wondered what our eyes' refresh rate is?

I wrote a paper on this topic (or just how digital imaging compares to our eyes and brain) back in college. I researched and researched and at the time (nearly 10 years ago) there just wasn't much to be found. There is just so much about the human brain and our nervous system that is still a complete mystery. One of the most fascinating pieces of information I dug up in my research though was a study done on the human eye and color. I can't recall the details of the conducted by who and when or where. But they used cards with color gradients in red, green, blue, white to black, yellow, purple, orange, etc.. In an attempt to determine just how many shades of color the human brain could differentiate. What they found is that it varied drastically from one person to the next and results for males fell within one range and female results were in a slightly different, but identifiable range. But the general consensus was that the human eye can differentiate over 2000 shades of grey, nearly 1500 shades of red, 2000 shades of green and 1200 shades of blue. Human vision is essentially 4-channels, Red, Green Blue and Grey. ...The study also found that a small percentage of individuals (just a few percet like 2-3%) of those tested had vision ranges that shifted slightly into the infrared portion of the spectrum. These individuals tended to be more sensitive to various "fluorescent" colors and could see more tonal ranges where most people just saw something white. They also found that many people are blind or not sensitive to certain tonal ranges. So they may be able to see levels in a red gradient up to a certain point and then several spaces on the chart would appear as a solid color tone before it began to look like a gradient again. It was fascinating, but I can't recall much more than that.

Our eyes also don't have a "frame rate" But it seems that our brain samples the rods and cones within our eyes individually and in a semi-random order with the ones near the center of our eyes firing more rapidly. I don't recall ever reading anywhere as to what the actual "sampling rate" or "refresh rate" of the rods/cones in our eyes actually is... Or if anyone truly even knows.

[Jared VanLeuven]

Fascinating stuff, man.

I guess they'll figure it out the day we get our movies delivered straight into the optic nerve. :)

[Jeff Kilgroe]

I'm waiting for the day I can dump my thoughts/dreams directly to physical media and edit them in real-time. :D

[ericyoung]

This came to my mind the other day as I was driving along and noticed the wheels of a car in the lane next to me. Its wheels (spokes) looked like they were moving backwards, as in the same effect that you can notice on film. I wondered what our eyes' refresh rate is?

Oh, and no, the wheels in question weren't those funky spinners... :)

Hmm. The only time I've seen this in real life is in the dusk/evening when the strobing streetlamps will cause this. What time of day did you see this?

[Jared VanLeuven]

It was daytime. I thought it was odd, myself. My first brush with that effect was as a kid watching "To Fly" at the Smithsonian Air & Space Museum in the scene with the old car. Funny how things like that stay with you.

[Anders Holck]

it seems that our brain samples the rods and cones within our eyes individually and in a semi-random order with the ones near the center of our eyes firing more rapidly.

I think it's the opposite: the center of the eye is higher resolution/Lower sensitivity/Low refresh while the outer region is low resolution/Higher sensitivity/High refresh.

Seems to be an inherited "Hunter" feature, which aids us to detect enemies coming from the sides of our vision.

[Jeff Kilgroe]

I've seen the reverse-spinning illusion with my own eyes here and there working with fast-spinning power tools and electric motors. I wouldn't think that car wheels would do it, but maybe they could. Perhaps it was an effect with the wheel design or spokes combined with turning at just the right speed.

[Jeff Kilgroe]

I think it's the opposite: the center of the eye is higher resolution/Lower sensitivity/Low refresh while the outer region is low resolution/Higher sensitivity/High refresh.

Could be... I was basically trying to recall from something long ago. I do know that our eyes are higher resolution (more rods/cones) in the center vs. the outer regions. The brain does process peripheral vision differently too and we tend to easily notice motion there rather quickly, but not notice as much for color or detail.

[gcaussade]

Interesting that you bring up seeing the wheels spin "backwards". I have seen this several times, but not often, but a friend of mine said that was impossible unless it was done on film. (or digital.)

You see it in car ads a lot, and now, when I've seen it in real life I've been telling myself that I must be hallucinating again! :)

-Jerry

[chuck colburn]

Interesting that you bring up seeing the wheels spin "backwards". I have seen this several times, but not often, but a friend of mine said that was impossible unless it was done on film. (or digital.)

You see it in car ads a lot, and now, when I've seen it in real life I've been telling myself that I must be hallucinating again! :)

-Jerry

Hi Jerry,

I was wondering also how Jeff was, and now you, are seeing car wheels that appear to be rotating at a different speed then in real time. Was the sun low in the sky and streaming through a picket fence? Or was there some kind or light reflecting back and forth between the two vehicles causing a strobing action? I can see it happening in a machine shop with crappy old flourescent lighting, but on the open road? Mayby it was due to that exotic mushroom souffle or a misspent youth! LOL

Chuck

[Kyle Mallory]

I know I've seen this (car wheels thing in person) multiple times myself... Damn you all, now I'm going to obsess about it on the way home from the office!

[Jared VanLeuven]

Kyle, let's rip it down I-15 and watch each others' wheels! Of course, we'll be at a crawl right about now, *sigh*.

[chuck colburn]

I know I've seen this (car wheels thing in person) multiple times myself... Damn you all, now I'm going to obsess about it on the way home from the office!

Is that the office you work at or the "office" we go to after work? Could make a difference.

[Don Woods]

I would say I would join you but thats a bit of a drive

[Steve Gibby]

Ah ha...the Beehive State Road Warrior Gang strikes again!

[jbeale]

I just looked at the wheels of some cars going by... if I blink, I can see a much-less-blurred version of the hubcap appear just briefly (persistence of vision effect?). If I blink rapidly, I can get a strobe-like effect. But I don't need to blink; if I watch a wheel and then suddenly look away, there is a similar brief freeze-frame effect at the moment I look away.

If you are driving and the road is slightly bumpy or the engine shakes, etc. I wonder if that vibration when combined with the persistence of vision effect could give you the backwards rotating wheels.

[Jared VanLeuven]

Well, didn't notice it on the drive home ... definitely made me weave a bit more than usual. :)

I wouldn't be surprised if it depended on a number of factors: speed, number of spokes, pattern of spokes (?)

Y'all drive safe now...

[Steve Gibby]

You can see the same effect on aircraft propellers. At certain speeds they appear to the eye to slow down and even reverse. The ability of the human eye to discern relative circular motion at certain speeds must be limited...

[Jared VanLeuven]

Yeah, come to think of it, I have seen it on propellers as well.

At the risk of going completely over the rails and down into Offtopic Ditch, here's a cool site with mind-bending optical illusions that illustrate our mind's wacky machinations:

http://www.michaelbach.de/ot/

[Steve Gibby]

Checked it out quickly...very cool.

I don't think it's off topic since as shooters, editors, etc. we all deal with images and the eyes ability to discern certain shapes, colors, and shades...

[Jeremy Hughes]

http://www.dvxuser.com/V6/showpost.php?p=686088&postcount=30

[Steve Gibby]

Yeah, come to think of it, I have seen it on propellers as well.

At the risk of going completely over the rails and down into Offtopic Ditch, here's a cool site with mind-bending optical illusions that illustrate our mind's wacky machinations:

http://www.michaelbach.de/ot/

Your wheel reversing phenonema is explained on one of the illusions on the link you put up. Link: http://www.michaelbach.de/ot/mot_spoke-illusion/index.html

[Steve Gibby]

http://www.dvxuser.com/V6/showpost.php?p=686088&postcount=30

Good catch...I remember that post of Graeme's.

[GlennChan]

"We do see aliasing on stuff that moves " Can't say I've seen that myself - care to elucidate? Is that spatial (which is what I'm on about) or temporal aliassing?
I was talking about spatial aliasing there. In real world situations, we generally don't see spatial aliasing on moving subjects.

To see for yourself, you can try printing out the following zone plate test pattern. You likely need to resize the image with "nearest neighbour" sampling in photoshop (or its equivalent in another program) so that it will print well.
http://www.worldserver.com/turk/opensource/z128.gif
Hold the printed version in your hand and warp the paper, or move the paper around.

A variation on the zone plate is as below.
http://glennchan.info/Proofs/forums/blue-zone.png
If you flick your eyes (or just move em around) between two spots on the zone plate, you'll see spurious sets of circles appear.

I believe the reason why this occurs is because our visual system can scan scenes through eye movements ("vernier scanning" is the phrase I'd use). Assuming that the target is stationary, this allows our visual system to get higher resolution and not be very prone to aliasing. This gels with experimental data, which shows that we can see alignment detail that is much smaller than our cones/photosites (hyperacuity on michael bach's site). From my own experiments with printing out a zone plate, it's also consistent since we see virtually no aliasing. This is especially so for the blue zone plate. Our eyes have three types of cones, one of which is very sensitive to blue (we can think of them as 'blue cones', which is somewhat accurate). These blue cones are only 2% of the total cones in the eye, and must be spaced with huge gaps in between them. These huge gaps will be very prone to aliasing- any detail that falls in between photosites can alias. The HVS avoids this by moving the eye, which means the photosites can scan the stuff that falls in between gaps.
Contrast this to camera systems, which don't do vernier scanning and are bounded by Nyquist/Shannon. They can't see detail smaller than a photosite, and they are prone to aliasing (unless heavy filtering is employed, which reduces resolution).

Vernier scanning should break down if the target isn't stationary. Our eyes can compensate for this by tracking the motion. However, for motion it cannot track, aliasing should occur. This (to my eyes) happens with the zone plate tests above.

2- Frame rate of the HVS:
Perhaps the frame rate of the HVS is flexible. In low light conditions, lowering the 'frame rate' will increase exposure time, which will increase sensitivity. This sort of gels with experimental data on flicker, since flicker sensitivity goes down as luminance decreases.

This might also explain why our eyes have less aliasing in dim conditions (when looking at a zone plate). Lower FPS = more motion blur = less fine detail that will alias.

[Jeff Kilgroe]

I was wondering also how Jeff was, and now you, are seeing car wheels that appear to be rotating at a different speed then in real time.

I haven't seen it with car wheels.. But as I did say, I've seen it with very fast spinning motors and power tools. It's easy to replicate indoors under flickering fluorescent lights, but I have seen it out in the mid-day sun too. Trying to keep your eye on a label of a cutting blade as it does 10Krpm or more can play tricks on your eyes. I suppose it could happen with car wheels if the conditions are right -- I have not seen this myself.

[J. Bernard Vallon]

I'm waiting for the day I can dump my thoughts/dreams directly to physical media and edit them in real-time. :D

I had a dream once in which I had a firewire port installed in my neck, matrix style. I could go to sleep (in my dream), and when i woke up, I could dump my dreams as a quicktime file and edit away. It was awesome, experimental-freaky footage.

When i woke up, without thinking i went to my computer (to dump the footage). Groggy, i got frustrated when i realized i had no footage to dump, and no hole in my neck.

::sigh:: One day....

[Jared VanLeuven]

Your wheel reversing phenonema is explained on one of the illusions on the link you put up. Link: http://www.michaelbach.de/ot/mot_spoke-illusion/index.html

Yeah, you're spot on with that, Gibby. Good catch there.

[Graeme Nattress]

Glenn, I have seen the effect you mentioned, and it nearly made me ill the first time. I'd got a new Epson printer, and there was a diagram in the manual with lots of diagonal lines on it. I'd moved the manual for a closer look and "wow" weird things happened. If you zoom in and out (manually by moving the manual) you get all sorts of weirdness.

So yes, we do get these effects, but it's rare.

All cameras really need the correct amount of low pass filtering - no escaping that one here.

Graeme

[Steve Gibby]

Graeme,

I think you'll get a kick out of some of those optical illusions on the link Pixelchef put up. Explanations are interesting too - they talk about aliasing, LPF, and things that are the kind of things that interest you. When you get a break between the whisky/coffee cycle, check 'em out.

Cheers!

[Andrew Benz]

....When you get a break between the whisky/coffee cycle, check 'em out.

LOL Gibby. Graeme, your results may vary wildly depending on which part of the cycle you decide to test these visual abstractions.:eek:

Cheers Guys.

[I Bloom]

Aliasing only happens when a regularly patterned sensor encounters a regularly patterned surface whose detail is finer grained than the sensor. In randomly detailed surfaces such as on living things alaising is never a problem.

There is one method of avoiding moire patterns that is seldom mentioned but my favorite. It's called stochastic oversampling and its the method used by Pixars renderman to avoid moire patterns on surface textures in CG. Basically if we were to downsample a 4K image to 2K we could, for example, use four pixels to compose one downsampled pixel, by say averaging all of their values. (the actual method is a bit more complicated) In stochastic sampling we would use only say three out of four pixels selected randomly. This additional randomness adds a tiny bit of noise to high detail areas but effectively eliminates Moire by removing any regular pattern in the sensor. Since red is a cinema camera there are also effective algorithms for temporarily ignoring pixels in 4K when interpolating the Bayer pattern. Since the eye is seeing many frames per second, there will be no perceived loss of detail in the moving image, while each individual frame might show some loss. In fact this is exactly what is happening when we look at projected film, of which each still frame does not appear nearly as detailed as the moving image.

You should be able to take advantage of these techniques on the RAW image but I would love to see such algorithms built into Red's firmware as I believe it leads to a more filmic image.

Cheers,
Ian Bloom
http://www.ianbloom.com
Red #657