Frame size comparison & red color banding?

[Can Scanner]

Here is a comparison of three frame sizes- 1080, 720, 480- each shot with either DVCPRO HD or DVCPRO 50.

http://www.greatmindspro.com/movies/Matt_St_Charles_HVX200_1920x1080.jpg

Notice the banding along the edges of the rose. It's much less pronounced in the 480 frame. Is this simply a result of DVC50 being less compressed?

Any feedback of this image is greatly appreciated. I intend to show it to potential clients (unless the general consensus is that it needs improvement).

Thanks!

EDIT: Per David Newman's advice, I converted new footage using the appropriate upres options with CineForm, and the results are very pleasing:

http://greatmindspro.com/movies/MSC_HVX200_Comparison_1920x1080.jpg

:thumbup:

 

[mactrix]

Actual it's no banding, these are sub-sampling and upscaling artefacts.

The chip size is about 1 Megapixel, so very close to 720p. For 1080i the picture is blown up by the image processor. Also the chrominance sub-sampling is now visible. Keep in mind that usually the image is not presented right in front of your nose. You have these kind of artefacts with many video formats but due to the distance to the projections system you won't notice them so much ...

 

[Verko]

Actual it's no banding, these are sub-sampling and upscaling artefacts.

The chip size is about 1 Megapixel, so very close to 720p. For 1080i the picture is blown up by the image processor. Also the chrominance sub-sampling is now visible. Keep in mind that usually the image is not presented right in front of your nose. You have these kind of artefacts with many video formats but due to the distance to the projections system you won't notice them so much ...

It is scaling artifacts, but not within the HVX. What you are seeing is low-quality horizontal stretching after the HVX, in both the 1080 and 720 images. The 1080 starts as 1080X1280 (native DVCPRO HD 1080 resolution), so to view it at the proper aspect ratio, something needs to stretch the image horizontally to 1920. Same with the 720; it starts as 720X960, and to display with correct aspect ratio it goes to 1280.

I don't know what platform you are playing with the image in, but if you look at the original frame natively (1080X1280, 720X960), you won't see the jagged edge. I edit in 1080 DVCPRO HD natively, so the only horizontal stretching that happens is for display; the actual file remains native. FCP's Best interpolation setting does a good job, however. Combustion's interpolater is good too; AE's is okay. Quicktime Pro does not do a good job scaling; Cleaner (on the Mac; I don't know about PC) does an excellent job.

The CCD''s in the HVX are 540X960 pixels, which make for 518400 pixels, not 1 megapixel. Everything shot by the HVX first goes to 1080P, and then from there gets scaled down in the HVX to the recording format. That's why 1080 definitely has more resolution than 720, even though the CCD size, even with pixel shift, is not high in pixels. 720, 480 both go through 1080 first. 480 benefits from oversampling.

I use the Kona LHe to output via HD-SDI through a Blackmagic HDlink to an Apple 23inch display, which shows 1080X1920 pixel for pixel. The Kona LHe is scaling the 1280 to 1920, and I can see a better version of the jaggies than in your file, but there are still jaggies. But that's only in the display. If I need true 1080X1920 for any reason, I scale to that with FCP best interpolation, or Cleaner, and everything is smooth.

Yes, there is chroma subsampling, but it's basically not visible and it's not responsible for what you are seeing in your image.

Respectfully,

Verko

 

[mactrix]

The CCD''s in the HVX are 540X960 pixels, which make for 518400 pixels, not 1 megapixel. Everything shot by the HVX first goes to 1080P, and then from there gets scaled down in the HVX to the recording format.

Wow, this is new for me. And it goes to 1080p50/60 too before scaling down to 720p50/60?

 

[DavidBeier]

The CCD''s in the HVX are 540X960 pixels, which make for 518400 pixels, not 1 megapixel. Everything shot by the HVX first goes to 1080P, and then from there gets scaled down in the HVX to the recording format. That's why 1080 definitely has more resolution than 720, even though the CCD size, even with pixel shift, is not high in pixels. 720, 480 both go through 1080 first. 480 benefits from oversampling.

Well, theoretically, the pixel shift is supposted to add 1.5 times resolution to the CCDs making the incoming signal 1440x810 according to Pansonic (so, if you believe in the power or pixel shift, the HVX200 has a bit more than a megapixel theoretically). These pixels are then uprezzed to full 1080p resolution (1920x1080). Then they are subsampled at either 1280x1080 for 1080p or 960x720 for 720p.

 

[Verko]

Wow, this is new for me. And it goes to 1080p50/60 too before scaling down to 720p50/60?

I don't know about the PAL version, but for the US version, as I understand it, everything goes to 1080 60 P, the camera does its processing, gamma, all that, and then it encodes the signal to it's final destination, whether that's 1080X1280, 720X960, or 480X720, DVCPRO HD, DVCPRO 50, DV, etc. I think it's clocking the CCD's at the true frame rate, whether that's 24, 60, whatever. If I remember correctly, its actually handling the image as full 1080X1920 at the DSP level.

That's why I shoot 1080 whenever I can, because the 720 has already started up at a higher level, and is downsampled from there. It's true that 480 DVCPRO 50 has a lower level of compression per pixel, so if you know you are staying SD, shooting in 480 makes sense, but the compression at 720 and 1080 is still mild and I would say irrelevant; the noise of the camera is way above any "noise" from compression. I do a lot of blue and green screen; everything keys beautifully and there are NO jaggies on the edges, even from the 4:2:2 chroma subsampling (except for the aforementioned display jaggies that aren't actually in the file).

I do put the Detail Coring at +7 for almost everything, as I only use Cine-D gamma and the noise must be punished. Then I add a little sharpening in FCP (usually +15 or so) after the keying plugin (not before of course). The same is true when I key with Keylight in AE; do the key, then color correct, then sharpen, all in one shot with the plugins in that order. That way the foreground or keyed element gets a little sharpness back, and the background doesn't get additional unnecessary sharpness.

By the way, if you calculate the data per pixel for 720 and 1080 DVCPRO HD, it's exactly the same. Remember that 1080 will only go up to 30 frames (60 fields) per second, where 720 is 60 frames per second; when you shoot 720 at 30 or 24, you aren't using the full 100 Mbits for the visible frames, even when not shooting in native mode. 1080 at 24 is really 1080 at 30 with a pulldown. So 1080 is always better, expecially for further post manipulation. The only advantage to 720 is less data to record.

Everyone, if I've got any part of the camera info wrong, speak up, but that's as I understand it.

Best,

Verko

 

[Verko]

Well, theoretically, the pixel shift is supposted to add 1.5 times resolution to the CCDs making the incoming signal 1440x810 according to Pansonic (so, if you believe in the power or pixel shift, the HVX200 has a bit more than a megapixel theoretically). These pixels are then uprezzed to full 1080p resolution (1920x1080). Then they are subsampled at either 1280x1080 for 1080p or 960x720 for 720p.

Yes, I believe in the power of pixel shifting! hehe I was just stating the actual pixel resolution of the CCD. As you say, even with the theoretical power of pixel shifting, it doesn't get up to the level of 1080. I was pointing out that even though the numbers don't lead to believe that the 1080 should be much better than the 720, there is a clear difference, especially with fine detail, and for me, it's worth staying with 1080. Everything goes through that vertical resolution anyway, so staying with 1080 means one fewer (vertical) scaling step. If you check out the link in the BBC trials P2 and Infinity thread, it shows clearly the difference between 1080 and 720 in fine detail.

I've found that going with the +7 coring with adding detail in post, even my low-light work has totally acceptable noise. I even add +6db gain if I need to; it just looks like in-between 500 and 800 ASA 35mm. (+12 is definitely too much 99% of the time). When going down to 480, like for DVD's, the noise just goes away, as it does with good 16mm. When I downrez to 480, I do add a very slight gaussian blur (applying it before scaling down) to prevent aliasing.

Anyway, back to work!

Verko

 

[DavidBeier]

Verko, would you care to post some pics of some of your 1080 stuff?

I do put the Detail Coring at +7 for almost everything, as I only use Cine-D gamma and the noise must be punished. Then I add a little sharpening in FCP (usually +15 or so) after the keying plugin (not before of course). The same is true when I key with Keylight in AE; do the key, then color correct, then sharpen, all in one shot with the plugins in that order. That way the foreground or keyed element gets a little sharpness back, and the background doesn't get additional unnecessary sharpness.

First off, if you're trying so much to reduce noise then why not use the less noisy Cinelike V? Aside from that, your setup is interestin and I'd really like to see some pics. People like Ash and Barry have indicated that +7 in coring can cost you high frequency detail so it strikes me as odd that, since you are so concerned with preserving detail, that you'd do this. What are your detail and V-detail level set to when you shoot? It also strikes me that adding detail in most isn't the best method. In camera, the affects are added before it is compressed in DVCProHD (so to the 4:4:4 1920x1080 image). Trying to add them in post would mean doing so to the compressed image so would theoretically not yeild as good results.

By the way, if you calculate the data per pixel for 720 and 1080 DVCPRO HD, it's exactly the same. Remember that 1080 will only go up to 30 frames (60 fields) per second, where 720 is 60 frames per second; when you shoot 720 at 30 or 24, you aren't using the full 100 Mbits for the visible frames, even when not shooting in native mode. 1080 at 24 is really 1080 at 30 with a pulldown. So 1080 is always better, expecially for further post manipulation. The only advantage to 720 is less data to record.

You're conclusion at the end of this statment is kind of confusing. Everything you were talkinag about made sense untill we got to "So 1080 is always better." How do you mean better? In terms of motion and certain kinds of image manipulation, I'd say 720p is much better since it handles pure progressive frames constantly. You can record at a silky smooth 60p which can't do in 1080 and you don't have to worry about pull down or redundant information (assuming you are working in Native mode).

 

[Verko]

David,

Thanks for your post. I'll try to reply more comprehensively and post some pics later, but quickly, about the 720 vs. 1080, if you are doing 60 fps, I like 720 better also. All my stuff is 24 or sometimes 30. I don't like interlacing. With 1080 24pA, yes, it requires conversion, but with the new P2 import in FCP, it's seamless (except the TC; see my post on that). I have been using Cinema Tools to reverse telecine; yes it's a pain and an extra step, but oh well. The new P2 import in FCP takes care of the pulldown issue.

I add sharpening in post because I can control the level based on seeing the scene at full resolution. I can add just a little or a lot, depending on the scene, and I can always go backward. Once it's applied in camera, there's no going back. And, applying AFTER keying with green screen makes for much better keys; you don't have to clamp down to get rid of the "enhancement" for the edge to be clean. Less noise in the edge too. The DVCPRO HD compression is good enough where I don't think it makes much difference whether it's added at the full-quality stage in the camera DSP or in post.

I have been setting vertical detail at +1 and detail at +1. If you look at the blue channel only close up, you can see the "enhancement" at that level, but it's acceptable to me. Then I add more in post, which especially helps eyes pop; I find mushy eyes to be the biggest sharpness problem.

I like the smoothness of the CineD; Cine V is good too, but I like to see more detail in the dark areas and I can always crush the CineD. It's just a personal preference. The CineV shows less noise than CineD because it is crushing the image more in the dark areas, where the noise resides. I also set the master pedestal at -2 for most things, which reduces noise in CineD. I put master pedestal at 0 for night or low light, as it keeps some detail at the bottom. As for the other gammas, I find they produce real banding in any smooth gradient-like surface. I think the "grain" of the CineD helps break up the banding, which for me is preferable. ILM stated that they added "grain" to the Star Wars 2 and 3 footage for the same reason. The downside of digital.

I decided to go with +7 coring while on my first greenscreen shoot. The talent had on dark pants, and the noise was very distracting; the noise went away with a full crush of the blacks, but that's not the look I wanted, especially as he was going to be composited on an off-white background; I figured if he was really there in the BG, there would be plenty of fill. I did test composites on set before we shot for real, as the camera and workflow was new for everyone. +7 coring was the only setting that was acceptable in this case, and adding sharpness in post took care of the increased mushyness (yes, that's now a word, hehe). Even with the added sharpness, it's still a smooth look versus the Canon, for example. Looks somewhat like 35mm to me, especially when on standard DVD.

Going with 1080 allows me to have the coring at +7 without losing too much resolution. It's the sweet spot for me, versus 720, based on tests. I still sometimes do a test shoot at +5 or +3 coring, just to prove to myself that I am getting the best out of the camera. The banding with the other gammas means I don't even consider them any more, and I can always punch up the contrast at any time. I use Color Finesse Synthetic Aperture in FCP and AE for most of my final color work, though the FCP 3way CC does a good job for simple corrections. Color Finesse gives almost unlimited control, but it is sometimes noisier than the FCP CC because the image is getting converted to RGB and then back into YUV, whereas the FCP CC stays native YUV.

Random points:

-1080 is pure progressive for 24 (once reversed) and 30; for 60 it is interlaced, and I prefer 720 for 60 work.

-Detail vs. noise is subjective; each person will choose differently. The best thing to do is methodically test each combination of settings to see what you get. Make sure to view in the best way possible; viewing in Quicktime or the on-screen viewer of FCP is much more noisy than with HD-SDI and has worse gradient banding; the HD-SDI is more accurate to the image. I'm trying to get the best detail for the least noise, and for me, I've drawn that line with +7 coring, 1080 P, adding some sharpness in post.

-Image manipulation will always be better with 1080, as there are more pixels to play with. Motion manipulation will be better with 720 as you can start with 60 real frames. I agree with you, David, on that.

-Green and blue screen makes any image problems much more noticeable, especially when you are composting onto a simple background, like a color wash. That means that noise is much more noticeable than it would be in a regular image, as the background doesn't really have noise (unless you want it to).

-I'll check with my clients for posting some screen shots. I can post tests, of course, but I like to show my best work, hehe, when possible. You've heard the saying, the difference between an amateur photographer and a professional is that the amateur takes 100 shots and shows them all, whereas the professional takes 100 and maybe shows one. Not trying to make excuses. It may be next week before I can post pictures; I have been procrastinating on a project by writing these posts!

-Keep in mind that my approach comes from my preference toward 24fps and a smooth image. Everyone is different. I continue to test, and for some things, I will choose differently.

Thanks for reading,

Verko

P.S. I've edited a few things in the five minutes since I first posted, just so you know.

 

[Verko]

Still procrastinating. Dammit!

I grabbed a few stills to show for now. Making them small enough to post here degraded them too much, so they are on my FTP server:

http://70.90.167.217/ and then login and pw below:

Login Name: DVXuser
Password: H2V4X9

Use the grey download arrow to the right of the file, or right-click etc. as usual. If you input the wrong login name or password more than twice in a minute, you will be blocked; let me know if this happens with your IP and I'll unblock.

They are at 1280X1080, which is the native res to show true quality. If you want to see them in the proper aspect ratio, stretch the horz to 1920 in Photoshop.

First, let me say regarding the original post, the oversaturated red on the flower IS showing the chroma subsampling; my bad for saying otherwise. The chroma is too strong; with normal chroma levels, the subsampling is very hard to see. But part of the problem is due to lower-quality horizontal scaling.

See Buzz01.tif on my FTP. The eyes have the same problem and the flower mentioned at the beginning of this post. It probably doesn't help that the camera saturation is set to +5. Color matrix is normal. Gain +6. Yes, this is a dark scene! Buzz01nosharp is the unsharpened version, straight from the camera through FCP. Sharpening brings the "grain" back up again, but the size of the "grain" is smaller than if there was no Detail Coring, and therefore more pleasing. That's the real advantage of going with +7 coring, as I can bring sharpness back, which increases grain but at a smaller size.

Interestingly, Preview on the Mac shows a worse edge on some colors than viewing in Photoshop, particularly the muted yellow and orange in the lower right.

HW_BLVD01.tif was a quickee handheld shot on Hollywood Blvd; no, I am not a Scientologist, hehe (just a disclaimer, Jarred). It has various elements of high frequency detail, which is why I included it. Again, "nosharp" is without post sharpening; all of these frames are at +1 detail and +1 vertical detail, +7 coring. And CineD, Chroma Saturation +5, color matrix Norm. Notice the difference in detail in the leaf areas, as well as the brick texture (very subtle there).

For IJ01.tif, notice the detail difference in the rock between the sharp and non-sharp version. The sky is totally blown out, of course

With Fire01.tif, you can see the edginess of the sharp version aroun the top of the helmet. It's less objectionable in motion. I used a frame where the highlight on the smoke (lower left) is blown out and loses detail; previous frames were still showing detail. Still looks good, I think. This was CineD. You can also see detail difference in the weeds and in the patch on the firefighter's shoulder, for example. This clip was shot handheld at full zoom; the stabilizer takes a moment to kick in, and then it's quite good. Remember, the camera detail (shown in the nosharp version) is already at +1 and +1, so it's not like I am surpressing it except for what is happening by putting coring to +7.

I just got the okay from one client to show some screen shots of some of our green screen stuff. I'll also gather some skin tone shots, which highlight the niceness of CineD. I'll try to dig up some 720 vs 1080 stuff too. By early next week, I think. Must . . . get . . . back . . . to . . . work!

Verko

Last thought - I love the Cine color matrix, especially for controlled lighting shoots. It actually seems to make things even softer, though. Nice for skin closeups. The cine color matrix does seem to do strange things to skies in outdoor shoots when I can't add fill for the foreground, so I go to Norm matrix and +5 chroma, which is where I usually leave it for run and gun.

P.S. The attached shot is an aborted upload; I thought the Jpeg compressor messed it up too much, but I don't know how to delete it.

 

[Can Scanner]

Verko, very informative stuff! Thank you.
I had converted this footage with CineForm (a visually lossless codec) without resizing the frames, so as I understand it, unless each image is displayed on the right type of monitor, all the pixels get in a tizzy and don't line up correctly. 720p will have a pixel aspect ratio of 1.333, 480 will have a PAR of .9, and I don't know what 1080 will have because Premiere isn't telling me. I had mixed them all together in one 1080 project (which was a CineForm preset with a frame size of 1440x1080, to further complicate things.)
Anyway, what did everyone think of the picture I posted?

 

[Verko]

Verko, very informative stuff! Thank you.
I had converted this footage with CineForm (a visually lossless codec) without resizing the frames, so as I understand it, unless each image is displayed on the right type of monitor, all the pixels get in a tizzy and don't line up correctly. 720p will have a pixel aspect ratio of 1.333, 480 will have a PAR of .9, and I don't know what 1080 will have because Premiere isn't telling me. I had mixed them all together in one 1080 project (which was a CineForm preset with a frame size of 1440x1080, to further complicate things.)
Anyway, what did everyone think of the picture I posted?

It's a nice shot, the DOF helps the flower stand out. The chroma is clipping in some areas though, losing detail. Nobody would notice that at first glance, IMO.

I am not familiar with the details of Cineform, but I believe the image is being resized as it appears to be the correct viewing aspect ratio. The original for all three look differently in the horizontal; 1080 and 720 would be too thin horizontally for viewing, and the 480 would be too fat horizontally. The Cineform preset you mentioned would also look thinner natively, but not as much as the 1280 of DVCPRO HD.

If you want to get rid of the edge jagginess of the rose, FCP has a filter called "Color Smoothing - 4:2:2" which will take care of it. Graeme Nattress (nattress.com) makes an even more intelligent filter to do the same thing. Both work wonders for green screen shot on DV. I don't know what's available for Cineform along these lines; does anybody know?

Back to the image, I wonder if the chroma is clipping in camera or in post?

Verko

 

[Verko]

For people downloading from my FTP, make sure to download Fire01.tif. It's the best shot of this group.

Verko

 

[Barry_Green]

Verko, feel free to procrastinate any time. Thank you for taking so much effort and going into so much detail with your posts!

 

[Verko]

Verko, feel free to procrastinate any time. Thank you for taking so much effort and going into so much detail with your posts!

Thanks Barry; coming from you, that means a lot. I've always appreciated your posts as well, and they helped me get to know the HVX much better than I otherwise would have! Same goes for everyone here.

Verko

 

[David Newman]

If you want to get rid of the edge jagginess of the rose, FCP has a filter called "Color Smoothing - 4:2:2" which will take care of it. Graeme Nattress (nattress.com) makes an even more intelligent filter to do the same thing. Both work wonders for green screen shot on DV. I don't know what's available for Cineform along these lines; does anybody know?

The conversion tool has the option to upres while it converts the MXF To an AVI, we do filter the chroma during this process, just like we do the HDV when up-converting 4:2:0 to 4:2:2. C.S. used the native DVCPRO-HD resolutions in his test then upscaled in Premiere Pro. C.S., Try the upres in HDLink -- you should be pleased, it is also fast. Although as you are using Aspect HD your conversion up scales are limited to 1440x1080 (1.33 pixel Aspect), Prospect HD gets you to1920x1080 square pixel.

 

[DavidBeier]

Verko, saw your pictures. First off let me tell you that's some really wonderful stuff. The colors, gamma, and compositions (most of what's really important in a picture) all look fantastic. That said, I really don't see a whole lot more detail in your sharpened ones compaired to the unsharpened. To my eye, both display the same kind of loss of high frequency detail that I see when I go ahead and bump the coring up to +7 (stuff like faces and skies look great but thinks like hair in a close up or grass in a long shot really suffer). I guess with all the 1/3 HD cameras, you really can't find a perfect solution. Either your willing to deal with noise and get some extra detail or your willing to soften up your image a bit so things look smooth and clean. It's just a limitation we have to deal with for getting features for 6k that used to cost 70k. Still, great looking stuff.

 

[Verko]

Verko, saw your pictures. First off let me tell you that's some really wonderful stuff. The colors, gamma, and compositions (most of what's really important in a picture) all look fantastic. That said, I really don't see a whole lot more detail in your sharpened ones compaired to the unsharpened. To my eye, both display the same kind of loss of high frequency detail that I see when I go ahead and bump the coring up to +7 (stuff like faces and skies look great but thinks like hair in a close up or grass in a long shot really suffer). I guess with all the 1/3 HD cameras, you really can't find a perfect solution. Either your willing to deal with noise and get some extra detail or your willing to soften up your image a bit so things look smooth and clean. It's just a limitation we have to deal with for getting features for 6k that used to cost 70k. Still, great looking stuff.

Thanks for the kind words, David. And you are exactly right; especially with a 1/3 inch HD camera for $6K, it's a tradeoff between noise and detail. We each draw that line where we like, and also depending on the project. I don't add a whole lot of sharpening back in, just enough to satisfy me. When seen on a large monitor, it actually makes a noticeable difference. At least with my monitoring solution, the trees and brick Scientology building on the Hollywood Blvd shot, for example, definitely are less mushy with some sharpness. I'm not talking about night and day; yes, it is a subtle difference, but I think it matters. And when I go to SD, if I have the coring down lower, the noise comes through enough to be an issue for me, while the +7 coring is great.

Try adding sharpness in Photoshop to the unsharpened version to see what different levels achieve. Sure, there is less detail to start with because of the coring. And compared to the other manufacturers, the HVX will have a smoother look anyway.

Anyway, glad you like the shots!

Verko

 

[Verko]

One thing I must mention, which has been said elsewhere - adding "sharpness" in camera or in post does not add detail, it just highlights it. Yes, more coring removes some detail, sharpness doesn't bring it back. It just highlights what is left. That's one reason I like 1080, because there is more detail than 720, and because I often set coring at +7, I need all I can get!

Verko

 

[Verko]

Follow up

Follow up

I know this discussion has drifted from the beginning of this thread. Sorry Can Scanner!

I wanted to add one thing: for me, the best reason for using +7 detail coring on greenscreen is that color suppression, or getting rid of background spill in the foreground, tends to increase noise in the foreground to a distracting level. I notice it especially around the hair on the side of the head. +7 detail coring completely eliminates this problem. I don't use cine color matrix because it just makes things way too soft, even though I like the color.

I did a green screen shoot last month where I put the detail coring at +5, just to revalidate my original tests. Sure enough, I had the excessively noisy hair on the side of the heads, where it's reflecting the background slightly. Tweaking the color suppression to as far as it would go until hitting skin tone mostly did the trick, but one guy had on a lighter blue shirt, and its color was close enough to green that I wasn't able to clean up the side hair noise without totally introducing noise in the shirt. Oh well. It was a job that allowed for some experimentation.

I could use another gamma that gives less noise, but I am addicted to the look of CineD. I love what it does with highlights especially, Adam Wilt's recent tests notwithstanding. And I find banding, or solarization, in smooth skin tones with the other gammas.

Can Scanner, have you tried David Newman's suggestions for Cineform? Please let us know. I'd love to hear and see how the different input flow would affect your shot.

Also, have you tried shooting with less chroma and/or a half-stop slower f-stop? I think the red was not only slightly oversaturated but also a little bright.

If your client is new to working with HD, I think they're gonna love it. More than half of my work is now HD through to final delivery, and everything I shoot is now HD. Largely due to the HVX and the pain-free workflow.

Verko