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 :)

OLPF's

B

Bill Anderson

Well-known member
Joined
Jun 8, 2007
Messages
1,378
Reaction score
0
Points
0
Location
Vancouver
Any thoughts on this: do away with camera OLPF and provide a filter (or filters of varying strengths) for times when one is needed to break up certain frequencies: perhaps even a variable dial in OLPF? Coupled with a type of "viewing glass" that reveals potentially problematic frequencies with a given sensor. Hold the viewing glass to up to your eye to check for conflicts, and filter from there ….. Then a high percentage of shots get to benefit from no (or minimal) OLPF…. Is this possible, practical?
 
The OLPF provides for IR filtering and protects the sensor. It's not the kind of thing you want to be changing on a regular basis. The best approach is the one we take - a good OLPF that removes the frequencies most likely to cause aliasing. The last thing we want to see are nasty aliases and moire in the finished movie and have people "blame" the camera because they don't understand the reasons behind proper optical filtration.

Graeme
 
Nice idea Bill, how do you propose to achieve it? You don't need a olpf to have Ir filtering or a sensor cover.

The problems come from interaction between light details and sensor pixel structures. Some can be removed, but unless you have a 100% fill factor, some olpf effect is desirable to at least bridge details heading towards gaps between sensor pixel pads to the actual pads. This will minimise the details in those spaces causing flashing as the image moves across them (as they fall in and out of pixel pads). Foveon has near 100% fill factor, and uses micro lens array to gather the light to the pixels, something Sony used to do with CCD's years ago. Bayer though, is a compromise, it is inferior and has a lot of issues to do with the compromises, tricks. Three chip and foveon are not perfect, and have higher data requirements, which led to Bayer becoming popular in cinema cameras. But because Bayer samples only one primary color per pixel, instead of the three primaries, so the other two have to be worked out or guessed per pixel. These gaps in the other colors also causes increased optical aberrations. To get over this a bloody great big olpf maybe used, to spread the light (including the missing light) out over neighbouring pixels, including ones that sense the missing colors. The bigger the spread of the olpf the more the contrast of the pixels is reduced, and the pixel data becomes muddled, as you basically are introducing blur. The smaller it is, the more you increase aberrations, aliasing, miore etc. That is my understanding of it anyway, Graeme can verify wat is right. I had a better post recently, but can't seem to find it now.
http://www.reduser.net/forum/showthread.php?98130-4K&p=1186177&highlight=#post1186177

The issue of being a bit sensor specific, means the eye glass has to match the effect of the sensor.

Mosaic makes aftermarket olpf mech to add to cameras that do not have them. I do not know how good they are, but better than nothing.

http://forum.blackmagicdesign.com/viewtopic.php?f=2&t=6824&start=40#p44816
 
Absolutely David. Whereas a bit of moire can be problematic in a still, you can theoretically go in by hand in Photoshop and carefully deal with it. Such fixes are nigh-on impossible in motion. I've seen shots from some moire-prone cameras look disastrous in the grading suite, and generally such issues may be hard to see while shooting too, making the "surprise" even more shocking when the moire appears.

Wayne: "but unless you have a 100% fill factor, some olpf effect is desirable " - even cameras with 100% fill factor will show issues from aliasing and moire, but in 3 chip cameras or Foveons the result is luma aliasing. A mild amount of chroma aliasing on a bayer pattern sensor can be removed reasonably easily in post without ruining the image, but any amount of luma aliasing is very hard and troublesome to remove, if it's at all possible to do so.

The best place for the OLPF is near the sensor, and as is pointed out, designed with the sensor in mind. Adding an after-market one in front of the lens can help for those cameras that lack one, but it's not optimal. OLPFs are also expensive precision components......

Graeme
 
Yeah, stills are a breeze. But I was wondering about the possibilities that might lie in future innovations regarding OLPF's and cinema use.


I understand that OLPF's must be very fine tuned and exacting in their positioning, and I certainly wouldn't mess with an OLPF in its current typically fixed position, but is this fixed idea likely to change?

Although Nikon D800E didn't exactly remove OLPF, but negated it, I've found it to be quite manageable; not to mention the footage I have seen with other higher end camera's OLPF's removed.


I'm not preaching any of this, and I know there can be a world of hurt out there for nonchalant use of this practice: I'm just curious about the ( sanctioned) possibilities.
 
Graeme, I have a request, Ill put it on the end.

Yes, that I suspected, but I do not want to go without olpf entirely, but see a combination active intelligent removal used with it, on camera or in post. As long as it covers out the spacing you should appreciatively reduce flickering from details hidden (but revealed over time and reinsertable in surrounding frames. This all next level future techniques.

An image is largely likely to be like past images, and in the image itself predictable. Because moire etc is caused by spatial placement of rigid structural component, their effects are largely predictable and filterable out of the image. An intelligent program should be able to see if there is a problem where there likely is a problem, identify it, and identify and restore original image data component. It nay hit be perfect, but as long as it is so close it looks good, it is good. What do you think Graeme?

Graeme, my research load makes it difficult to explore everything to get a proper handle on it, and I'm veering out of cameras to other areas. So, I would like to ask, could you please explain to me what you are meaning with the 100% fill factor and luma moire issues? As far as I see, true 100% fillfactor should naturally anti-alaise itself, as any detail intersection between multiple pixels should evenly geometrically divide and proportion the image detail among the pixels leading to smoother transitions. I know with fillfactor solutions that microlens arrays can introduce aberrations into the light, but in itself is not indicative of 100% fillfactor but maybe the regular experience on many large fillfactor sensors, as not many have a true 100% sensor pads (and physically virtually cannot due to need for insulation between sensors. Thanks.
 
I think I have seen people remove it, wondering what sort of suction cap I need ;)
 
With the SI2K there is a mention in the manual that this camera can be operated with the OLPF removed provided an IR panel is substituted. I tried instead in a situation lit with discharge lamps to eliminate IR.

From a practical standpoint, I did not see sufficient improvement in apparent sharpness to outweigh the risk of edge twitter and aliasing without OLPF that I could not achieve by sharpening in post.

Shooting through a groundglass 35mm adaptor is a little different. An apparent 15 to 23 points of improved sharpness was yielded, depending on the lens on front of the adaptor and its iris setting. The groundglass to some degree seems to act similarly to an OLPF panel but there was an apparent twittering on a bright sharp edge, even through the groundglass.

This groundglass stuff is of course irrelevent to Red users unless you are thinking of making a "coatwolf" rig to do large format or plate camera image relay, or 35mm to 35mm for a full non-coherent optical effect.

There is another wrinkle to the business of removing the OLPF. Your backfocus will change, maybe outside of the range of available adjustment if your lenses are already a bit off.


I would recommend you DO NOT remove the OPLF at all unless you are filthy rich and it doesn't matter that you may have to pay the man if you can't get it right again. If I cannot see enough realworld justification with a 2K camera, there is even much less reason to do it with 4K or 5K. The chances of getting some crappenstance on the sensor or the rear face of the OLPF and not seeing it until you go shoot something are just too great. They are a dog to keep dust and dandruff off unless you have a proper cleanroom environment to do the job in. Getting my OLPF rear face clean from just one speck of dust was a major mission and that was with the benefit of having a CRT TV operating close by to draw the dust away.

There are special techniques the optical engineers know.


Companies which make things for sale, generally get as much performance out as they can whilst preserving reliability and have had more resources than a set of screwdrivers, close-up glasses, a ringlight magnifier and a kitchen table. Just because you think you can make it go faster, higher, wider, smellier or whatever, does not mean that you should. Something else usually has to give way as a compromise.


A rubber sucker may leave a mark on the OLPF which may be a dog to clean off. Like some automotive wiper blades, some rubber suckers may have microscopic hard abrasive particles entrained in the rubber which may mark the optical surface.


All that said, if you are competent and have the right tools to do it, why not but to re-iterate, I wouldn't bother.
 
Last edited:
The problem with aliasing is the "fold back" / "mirror" nature of the frequency distribution of the artifact induced. This means that high frequencies can lead to low frequency artifacts, and they're essentially un-removable due to their frequency now being at the same frequency as real image data. Indeed, that's why they're called "aliases". So Wayne, even if you know the frequencies, you can't remove them without removing actual image data too.

So moire/aliasing is caused by sampling the image. Filtering needs to be applied to stop all frequencies above that which will alias from being sampled. 100% fill factor basically changes point samples to area samples (and any real world sensor is going to be somewhere in-between point and area, not least due to use of micro-lenses). Being an area sample doesn't filter the frequencies per-se. A point sample will increase the frequency content of the image though, making aliasing worse, but even with area samples and 100% fill an image can and will still alias.

Graeme
 
Robert, I was joking. Think about how funny it would look seeing somebody try something like that, that's the joke. I doubt it would work anyway, they should not be easy to get. off. I would be throwing the olpf away anyway.

You are right, some 35mm adapters give a nice old film look and diffusion. Look for Marla the movie, one of the early adapter short films, very well done.
 
Graeme Nattress said:
The problem with aliasing is the "fold back" / "mirror" nature of the frequency distribution of the artifact induced. This means that high frequencies can lead to low frequency artifacts, and they're essentially un-removable due to their frequency now being at the same frequency as real image data. Indeed, that's why they're called "aliases". So Wayne, even if you know the frequencies, you can't remove them without removing actual image data too.

So moire/aliasing is caused by sampling the image. Filtering needs to be applied to stop all frequencies above that which will alias from being sampled. 100% fill factor basically changes point samples to area samples (and any real world sensor is going to be somewhere in-between point and area, not least due to use of micro-lenses). Being an area sample doesn't filter the frequencies per-se. A point sample will increase the frequency content of the image though, making aliasing worse, but even with area samples and 100% fill an image can and will still alias.

Graeme
Click to expand...

But how will it alaise Graeme? A genuine theoretical 100% fill factor sensor does not need micro lens, or have other sensor structures except its surface and it's boundary to the next pad, where would the frequencies come from, what would produce them? we are assuming all sensor pad sensor front, and circuits on the back of the pads underneath.

When In look deep, I look broad and deep.

Thanks again.
 
It will alias because all sampled systems alias if fed too high a frequency content. And as many people have seen luma aliasing issues on Foveon sensors, reality fits with sampling theory. Fill factor basically effects sensor MTF. Low fill factor approximates point sampling and very high MTF, 100% fill factor is an area sample with lower (but still good) MTF. A high MTF will mean you need to do more work to reduce frequencies that alias, and a low MTF means less work (but still work, as the MTF isn't zero) to reduce those frequencies that cause aliasing. Indeed, you've really got to think that a high fill factor doesn't lower MTF, but that a small fill factor does increase MTF and make harder work for the optical filtering necessary to reduce aliasing.

Graeme
 
Thanks, but how does the higher frequencies physically produce aliases on sampled systems? Doesn't lens primarily produce mtf in the first place, Graeme, how would dealing with mtf their affect mtf on the sensor?

Thanks again.
 
Each and every part of the imaging system has an effect on MTF - lens, OLPF, sensor. MTFs are multiplicative, which means at best they can only pass through the MTF of the previous part of the system, and in all practicality will reduce it. For more on aliasing try this: http://www.red.com/learn/red-101/resolution-aliasing-motion-capture

Graeme
 
I should also ask, if having a pixel pad size with 100% fill factor, large enough that all light hitting the boundary takes up less than the lowest quantity of the sampling depth (1 bit) would virtually eliminate aliasing.

In forgot, In think we were both talking about different aspects. Also, through software you should still be able to be restore image data. The effects described effect the accuracy of the pixel rather than what "should" be there. I came up with ideas for restoring image data long before commercial auto computer systems for restoring old film. So you can see where I was coming from. But back to the sampling theory topic.

I the end I'll probably figure out some minimumalistic system for dealing with the effects that can't be adequately dealt with. I'm basically just only against the pixel diluting affects of olpf, rather than their use at all. Many years ago I had ideas on how to deal with the effects on images of poor optics, noise etc that industry is catching up to. I identified that you could make a small camera produce much better pro video images, which gave great flexibility and ability to use types of optics you would not use normally.
 
I think using time and motion alleviates some of the concerns with proper image processing.
It's a motion picture camera.
 
That was my point in the 90's.
 
I made my point with this in my "got 6K ..." thread.
And I implemented it.
Now my processing is more advanced than the old examples in that thread, except for the last 6K still from India that I posed a link from .

Stay tuned for my little short film. All genuine 4K .

Wayne Morellini said:
That was my point in the 90's.
Click to expand...
 
Graeme Nattress said:
Each and every part of the imaging system has an effect on MTF - lens, OLPF, sensor. MTFs are multiplicative, which means at best they can only pass through the MTF of the previous part of the system, and in all practicality will reduce it. For more on aliasing try this: http://www.red.com/learn/red-101/resolution-aliasing-motion-capture

Graeme
Click to expand...

Graeme, is that the only thing you mean? In know that, the interpretation of it though is different. Indeed a point sampling mechanism and particularly bayered, due to the way it works, is going to have trouble with that, and I have avoided cameras, and ones with weird corner shaped, and position transposed, straight edges from these effects with peculiar binning (be cautious of good spec sensors on really cheap cameras). But at 100% fill, sensor edge to edge, the details are going to naturally blend into each pixel area with some refractive interactions and across pad boundaries. The true detail of the scene as it crosses a pixel represents the realistic average value for that area, and you expect it to naturally vary with content realistically. On a point sample, particularly bayer, this could make the pixels flash like crazy, and it would produce strange patterns and colors, with 3layer on the 100% pixel you expect it to naturally drift with the underlying detail. But, foveon images were not representative of a true 100 percent fillfactor sensor, just nearly, and the microlens used to archive that have their own issues.

Going back to an theoretical example, true 100% fill true 100 accurate 3 color sample across it, using a high f rate aperture to remove certain effects. What affects should I see on the motion images, any other causes, and how annoying should the example from the link be on this setup? Just trying to see a way past the problem Graeme. One thing I've learned from history, in that experience can be a poor predictor of what might be done new. Maybe we will cone up with a new solution for this? One thought, is that on 100% fill you could just increase the resolution, like res does, but to blend adjacent pixels. In am not really for that either, because as you move across the screen the amount of camera pixels blended into display pixels will vary, with some pixels on a theoretical system for example purposes, with same vertical and greater horizontal res, being a blend of two adjacent camera pixels and having no true data of it's own, dropping the resolution of that column. Probably why 720p tv's were sometimes 1366 wide. Like what Is said Song might be doing to get 8k artificially from there sensor. In also see how this might help bayer to get node than one color sample in a pixel.

I was thinking of trying to see how 8k looked unbayered one day from a camera, as it is so high res to blend. But after considering this here, I think I might reconsider. With some of the techniques I am considering I have figured out I can get most of the advantage of a bayer system from a 3 color system instead anyway. One day I might be able to give it a try. Waiting for chips. But I say to people, if you need it now, can easily afford it, buy what good you can get now. Red delivers good results.

Thanks Graeme?
 
You must log in or register to reply here.
Back
Top