Noise in green foliage

[Shekar Dattatri]

Hi, I have a peculiar problem common to both my XL-1 and XL-2 (PAL). Whenever I shoot a wide scene which includes green foliage in bright sunlight, the noise level in the green is atrocious. The rest of the scene is fine, only the greens are a problem and only in bright light and in wide shots. But since I am a wildlife filmmaker and tend to shoot a lot of green canopied shots, this is a major problem for me and I would very much like to minimize or eliminate this. I was hoping that the XL-2 would not have this problem, but alas, such is not the case....or so it seems. Any advice would be gratefully received!

[Pete Bauer]

There is a long thread from a few months ago that describes artifacting with fine detail that MAY be of some help:

http://www.dvinfo.net/conf/showthrea...s&pagenumber=1

I'm not too knowledgeable about this, but I have a suspicion that what is going on is that the fine detail is causing some combination of moire effect and/or DCT compression artifacting. If so, it isn't really a problem specifically with the camera; it is just the limits of mini-DV technology and you'll have to experiment a little for best results.

Changing your coring setting may help. Since you have a separate post asking specifically about coring, I'll post in that thread a short description that comes from the Canon web site, (which will then represent the sum total of my knowledge coring!). Perhaps reducing the greens during your shoot and boosting them again in post will help get better compression -- don't know if that'll actually work, though. And it may be that you'll have to zoom in or out a bit to avoid the "sour spot" where the amount of detail in the frame causes the most artifacting.

[Kevin Kocak]

I have had problems with the exact same issue. I have also posted about this and gone a couple of rounds with Canon about this. After much testing I have found some settings to change that my help a bit. There is the coring setting as mentioned above. You can also play with Noise Reduction and Sharpness. If you are shooting in progressive mode (24p or 30p) you can lower the vertical detail. These help a bit but they can only do so much. I first noticed my issue while trying to do a swooping shot on a jib across a garden. The leaves and grass were dancing away. The movement was very slow so it wasn't that. The lighting was fine so it wasn't that. Everybody seemed to have an excuse for it. Basically it is just a limitation of the medium. If you are goig to be shooting a lot of stuff outdoors in rich foilage you might want to re-think your camera choice. Also try some side by side tests with other cameras if you can. See if there is one that works better in those conditions.

[A. J. deLange]

The DCT compression algorithms work by transforming the image into the frequency domain, throwing out information in the high frequency parts of the spectrum (this means using fewer bits to represent the high frequency components or put another way quantizing them more coarsely) and then transmitting or storing the processed data stream which, as bits have been thrown away, is compressed relative to the raw data. This often works very well because most images are highly spatially correlated (another way of saying they don't have much high frequency content) which is the basic reason it is possible to do compression. Foliage shot at a distance is bad news for a DCT algorithm, especially in video. To begin with the largest component of luminance, the channel which carries all detail information, is green and foliage is green. Second, obviously, a scene with leaves at a distance has lots of detail, lots of high frequency component or little spatial correlation depending on the terminology you prefer. When the DCT is taken the high freqency "bins" contain as much or nearly as much energy as the low frequency bins. In a still camera the algorithm recognizes this and decides that it can't quantize the high frequency bins more coarsely than the low frequency ones and produces a relatively large JPEG file - the compression ratio is not as great as it is with a more normal scene but the result looks OK. In DV this option is not available. 25 mbps is the limit and bits MUST be thrown away to make the stream fit the channel. When the image is reconstructed from the reduced bit stream the result is distortion in the high frequencies. This distortion appears as numerous areas of a couple of pixels which are at the wrong level. This is similar to the way noise looks but I think if you look closely you will decide that what you are seeing isn't quite the same as noise.

All the above assumes that the leaves are large (close) enough to be below the basic resolution of the sensor. If that is not the case i.e. the leaves are near pixel size on the CCD you will have garden variety aliasing to deal with. Or you may have a combination of aliasing and DCT distortion.

I don't think there is much you can do about this. Minimizing the percentage of the area which is in foliage may help if the algorithms are sophisticated enough to allocate the bits saved in smoother areas to the foliage areas and aliasing isn't the major cause. Using a softening filter will lower the demands on the system (and reduce aliasing) but the leaves will now be blurred. If that's preferrable to the distortion the problem is solved. Getting closer to the leaves will reduce the high frequency component but that may not be possible. Coring may work but if you over do it you can get the plastic look I mentioned in the related thread.

I hope this is of some help I know this can't be very easy for the non engineer to understand.

[Shekar Dattatri]

Kevin, thanks for sharing! Misery loves company, and I'm glad I'm not the only one to have experienced this problem! It could well be that this is a limitation of mini-DV. I guess we must endure what we cannot cure, and try and work around the problem.

AJ, many thanks for that scholarly treatise on the subject. I'm no engineer, but I get the drift of what you're saying. And thanks for the caution on tweaking up the coring too much.

[Stefan Scherperel]

Here is a suggestion, I don't know if it will work, but you can always try.

I have owned a DVX100a for over a year, and one thing that I noticed now and then was when I was shooting in daylight, green shadows especially were noisy. It wasn't shot ruining bad, but something that bugged me. What I also noticed was that when shooting using tungsten balanced lights, noise in shadows was non existant, even in greens. So after some reading online I decided to give the 85B (daylight to tungsten) filter a try. This is an orange filter that balances daylight to use with tungsten settings. So when I am outdoors, I keep the color temp set to 32k or white balance through the 85B filter. I have noticed that shadows outdoors are no longer noisy.
The science behind this is that the CCD's are natively balanced closer to 32k, so when you white balance to 56k you are boosting the blue CCD substancially. However, if you optically switch the white balance so that you can shoot outdoors and still have the CCD's close to 32k, nothing is being boosted. I have used this meothod with Digital SLR's as well and have noticed similar results.
I'm not sure if it is the same problem, but you could give it a try. Beats buying a whole new camera.

[Shekar Dattatri]

Stefan, that's an ingenious solution and so easy to try. I shall certainly give it a shot and if it solves the problem, build a shrine for you in my equipment room!

[Stefan Scherperel]

sounds good to me!
Good luck, I hope it works out for you.

[Phil Rogerson]

Please write to us of your success, or otherwise. I'm seriously considering and XL2 for wildlife, and glad I read of this issue before buying. Thanks heaps.

[Shekar Dattatri]

I have done further tests today, but before I give the results, let me just clarify a few things. The 'noise' I referred to in my original post isn't, in the strictest technical sense, noise (as in 'grain'). I used the term loosely for want of a better description. It is some kind of unpleasant looking 'crawling' seen in green foliage in bright light, particularly when panning or tilting at wide angle. Close-ups are fine. I have since fine-tuned the camera settings. While this has not eliminated the problem, it has reduced it by about 50%. I have not yet experimented with using a 85 B Colour Correction filter as recently recommended by one of the forum members.

When I first used the XL-2 (at factory default settings), I was horrified by the 'ultra-sharp video look' of the images, which was so different from what I was used to in my XL-1. I have now set my camera to Cine Gamma and Cine Colour Matrix and have reduced sharpness by -3 and am very pleased with the picture I'm getting. The above mentioned 'noise in green foliage' is considerably reduced with this setting. I also played around with the coring setting, but this did not seem to make a visible difference. I will try my current setting with the 85 B filter as soon as I am able to and post the results here. Another clarification I have to make is that I am not viewing the results of my tests on a studio monitor but on a 29" flat screen Sony Wega DRC television.

The 'noise in green foliage' quibble apart, the XL-2 is a fabulous camera. For anyone seriously considering wildlife filming, there isn't another camera anywhere near this price range that can do what this camera can do. As for reliability, I worked my XL-1 really hard for three and a half years in diverse conditions - sea shore, rainforest and the blast furnace-like 45 C heat of a Central Indian summer - and it never let me down. Only time will tell if the XL-2 is equally rugged and reliable. But in terms of features and its ability to customize picture quality, this camera takes my breath away!

[Marty Hudzik]

As a simple test you could try filming the outside foilage using the indoor WB preset and then color correct it in post. The theory proposed is that the chips are balanced for 3200 and outside is going to be more like 5600. So the camera is increasing color gain to compensate for the lack of warmth (5600 is much cooler temp) essentially adding yellows to warm up the image.

I would at least attempt this in post before investing in a 85 b filter. IT might get you headed in the right direction.

I have read somewhere where some videographers purposely shoot only in indoor setting and color correct in post for everything as it delivers a cleaner image. FWIW.

[Shekar Dattatri]

Many thanks for that thoughtful suggestion. However, being also a film cameraman (super 16mm) I have several sizes of 85 filters in my equipment store. I just need to resurrect them from wherever Ive stashed them, as I have not had to use 85 filters for a long time (since the advent of superb daylight balanced neg stock).

[Stefan Scherperel]

Hey Marty, your theory would definatly work, however, DV is very compressed, and changing the color that drastically would introduce extreme amounts of grain, artifacts and posterization. Essentially you are doing what the camera is doing, boosting the blue channel, but instead of doing it before compression like the camera does, you are now doing it after compression. IF the camera shot raw 4:4:4: it could defintly be done, but not with DV.
Shekar, it sounds like what you are experiencing is indeed a moire effect so I don't know how much the filter would take away. but indeed, if you have the filters on you it is definatly worth a try. As far as the DVX goes, it is the number one way I have found to elimintate grain altogether, but the XL2 is another story. Good luck.

[Marty Hudzik]

Stefan,
While I agree with you in general about DV compression, I am not convinced that a simple color correction introduces that much noise. It might in certain circumstances but not too much. I guess it depends on how noisy the camera boost is versus the software gain. Maybe this method held true back in the analog days when the chips on the camera added mucho noise in an effort to color balance. Anyway it was worth a shot.

Marty

[A. J. deLange]

Shekar,

If the "crawling" is active and shows no pattern when the camera is held still and the leaves are not being moved by the breeze then what you are seeing is noise (thermal, from the chips). If the crawling is only seen when the camera moves or the leaves move but stops crawling when motion stops but still shows no perceptible pattern then it is aliasing. If the disturbed parts look like square tiles several pixels on an edge and move about with the motion of leaves or camera then it is compression (DCT) artifacts that you are looking at. It is, of course, possible that you have all three at once!

I have to say that I don't much like the idea of shooting with the balance artificially set incorrectly. The reason for this is that color balancing is a simple matter of sligthly offsetting the gains in the R, G and B channels before going on to do luma and color difference computation. This is best done in the camera because you have at least 12 (or it may be even 14 - I don't recall) arithmetic in the camera. Once converted to DV you only have 8 and on average it's less than that because of the compression. The color balancing done by decompressing, color correcting and recompression is much more likely to introduce artifacts than doing it before any compression has been done.

I also think the reasoning given is backwards. Silicon based CCD's have less sensitivity at shorter (blue) wavelengths than long (red). Thus less of a blue (the noisiest channel) gain boost is required in daylight than in lower color temperature light. If you put on an orange filter you are blocking light to the least sensitive channel and will have to (effectively) boost it's gain more in post (thus bringing up its noise) than if you shot with sunlight or, better yet in this regard, sky light.

That's how I see it in theory but please go ahead and do the experiment anyway. It is often the case that the practice doesn't show what the theory predicts. Then we modify the theory!