New Sample Footage

[Steve Mullen]

Originally Posted by Brett Sherman

I hooked it up to my studio monitor and switched between Progressive and interlaced. The "watercolor" effect definitely appears in progressive.

You are using an HD monitor connected via HDMI? Which monitor?


What's interesting is that Sony did try to do "something" with V1E. So obviously they feel something can be adjusted. Which raises the question -- are there unit to unit variations. This was definitely the case with the JVC HD100 -- until they worked out a new QC procedure.

And, beyond the aliasing settings, there are the chroma filter settings.

And, while there are not 2 Encoders, the output stream should receive different flags. A P stream should be compressed as a FRAME and receive a FRAME-FLAG. This eliminates the concern that P is encoded as fields. It need not be. This is part of the MPEG-2 spec. Did Sony do this?

In the real-world of watching HD video -- I've got 3 hours of 24p from India and I haven't noticed anything different than my 4 hours of 60i I shot in the USA. But then I'm not freezing frames and magnifying to look for problems. And, I've posted several sets of I vs P captures and so far no one has noticed anything other than the P is softer -- which makes sense because P was shot at 5 and I was at 7.

I've been shooting DV and HDV for day 1 and there are always spots where things -- if you look for things wrong -- you'll find them. This is particularly true of HDV.

[Mikko Lopponen]

Originally Posted by Steve Mullen

And, while there are not 2 Encoders, the output stream should receive different flags. A P stream should be compressed as a FRAME and receive a FRAME-FLAG. This eliminates the concern that P is encoded as fields. It need not be. This is part of the MPEG-2 spec. Did Sony do this?

I'm guessing a bit here, but eventhough mpeg2 has lots of options for encoding, hdv is pretty locked down. That's why Sony doesn't use frame compression in progressive mode, but canon does. It's not exactly in the hdv-specs and so there might be compatibility issues.

[Mikko Lopponen]

Originally Posted by Brett Sherman

Inferior scaling both on the computer and with an HD display will add a watercolor effect that doesn't exist in the raw footage.

Scaling can do many things, but no scaling will make the picture look like those the were in the 25p thread.

And yes, that church show has a small amount of "smart smoothing" applied giving it a small watercolor like quality. This noise reduction technique is the source of the problems, but I have no idea why it can't be fixed or why the ntsc version doesn't suffer as much as the pal.

[Brett Sherman]

I've uploaded an AB comparison between interlaced and progressive. Go to www.geekstudios.com/demos to download the m2t file. I also uploaded JPEGs that illustrate the differences too. Both were shot with the same profile - sharpness at 7.

The progressive is softer and slightly more noisy (especially at the bottom left picture). I think I overreacted in my last post. I don't think the progressive performance is a deal killer for me. And it is definitely not as bad as the V1Es. I looked at the footage I shot in Nicaragua and am pleased with the results. There are instances where parts of the shots look like too much noise filtering is going on, but to me it doesn't ruin the picture. I'm very pleased with the dynamic range of the camera, it beats my old DVX-100 hands down. I don't know how it compares to the Canon A1 in this respect. There are scenes with bright light and dark areas together and it doesn't blow out the high areas and you still see detail in the dark areas.

I think the paint effect seems to be scene specific. Some exhibit it more than others. I think it's some sort of noise filtering going on. I wish Sony had done a better job with it. Hopefully, there will be some firmware update to improve it. The question for me is, should I shoot in Progressive or Interlaced? Interlaced is slightly cleaner, but Progressive has that film-look.

As far as monitoring is concerned. I use a Sony LMD-232 hooked up with HD component. While HDMI might be better, I'd rather be viewing on a professional monitor like the LMD-232. I'm unaware of any professional monitor with HDMI capabilities. You'd need an HDMI to HD-SDI converter which I don't think exist yet.

[Steve Mullen]

Originally Posted by Mikko Lopponen

I'm guessing a bit here, but even though mpeg2 has lots of options for encoding, hdv is pretty locked down. That's why Sony doesn't use frame compression in progressive mode, but canon does.

You'll note I said Sony "should have" used FRAME encoding. This is my opinion and we have no information on what Sony does -- although I've asked. Since Sony is a co-developer of HDV I assume it could do anything it wants as long as JVC agreed. But, even if it could -- I'm not claiming it does.

But, let's assume Sony does use FRAME encoding. Software encoders would have to check for a FRAME flag. Hardware encoders likely do so because they typically are more strictly imlemented. I've asked DSE if Vegas supports FRAME decoding, but never received an answer. DSE?

We are now way off in the land of speculation. But what the heck.

An oil or water color effect sounds to me like the loss of vertical chroma resolution. A vertical blur. Blur is the classic side-effect of chroma digital noise reduction (DNR).

DNR is done by taking 2 or 3 lines and averaging them. So which lines get averaged? And, is the chroma DNE done prior to recording or upon playback?

If done prior to recording, all lines would come from the same frame. Thus each interlaced field will have it's own chroma DNR. And, each progressive frame will have its own chroma DNR. I don't see how I and P could be different.


On playback, with interlaced video the lines SHOULD come the from the current field and the previous/next fields. This is 3D DNR. But, 3D DNR is not often implemented. Thus, all lines come from the current field. These lines are actually not spatially close -- so the image is blurred vertically.

With progressive video -- all the lines are within the current frame, so all lines are spatially close. Thus the image should NOT be blurred vertically. Whoops! Exactly the reverse of what folks claim to see.

But, we are dealing with 4:2:0 and (perhaps) FIELD encoded or FIELD decoded progressive video. 4:2:0 sampling spreads-out the chroma lines very differently than 4:2:2 and 4:1:1. So intutively I can see why chroma DNR could cause a problem. But, it should be equal for I and P.

But, if progressive video's odd and even lines are encoded OR decoded separately, then when a frame is assembled--although all lines are spatially close, the odd lines and even lines may be "different." So averaging them could make a bit of a chroma mess. Thus, I can see how chroma DNR could create vertical blur on P.

But here's where it gets interesting. If Sony doesn't use FRAME encoding, then the decoder makes no difference. But, if Sony does use FRAME encoding -- it is critical that the decoder check for a FRAME flag.

It's very likely the software decoders folks are using do NOT check the FRAME flag -- hence all these frame grabs that look bad. But, if the V1 has a proper hardware decoder -- those of us who are watching video will not see a problem.

Note -- I'm not talking about the V1E.

[Alex Leith]

Thanks for that interesting theory Steve.

Would it still be possible to have "frame" encoding across progressive segmented frames, or are the two mutually exclusive?

[Alex Leith]

Looking closely at the (non hardware decoded) images, it almost looks like excessive coring within macroblocks which don't show enough variation in tonal range.

[Piotr Wozniacki]

Alex, could you please explain the term 'coring' to me? THX!

[Alex Leith]

As I understand, it's a noise reduction technique where a pixel of one colour surrounded by pixels of a similar (but different) colour is "cored" out and replaced by the second colour. It reduces grain noise and can be used to smooth skin or sky detail. If it's used too heavily then it loses fine detail. Usually there is a threshold set so it won't smooth colours that are too dissimilar.

Looking at the image of the church and the other comparative images that Brett posed, it would appear that the paint effect happens in areas where there are similar tonal ranges. In the workshop image on Brett's web page, you can see it (comparatively) on the foreground handrail - in the interlaced image there is fine detail on the moss; in the progressive image the detail is gone where colours are similar, but retained across colours which are dissimilar.

It also looks like this effect extends to the edges of the compression macroblocks, as there are evident (16 x 16?) blocks of colour, where once there was fine detail.

[Alex Leith]

Further there seems to be added detail noise (crawling edges) on top of the smoothing.

[Alex Leith]

In fact, those solid macroblocks change from frame to frame, which seems to cause what could almost be described as a "shimmering" effect around high contrast edges - edges that aren't cored out.

Is that the crawling ants?

[Mikko Lopponen]

It looks to me a bit like this smoother:

http://neuron2.net/hiq/smoothhiq.html

an extreme example:

http://nickyguides.digital-digest.co...omer-noise.jpg

The smoothing isn't as bad in the v1e, but it's still something that should not be active unless there is excess grain like in that example pic because it will lose detail.

[Mikko Lopponen]

Originally Posted by Steve Mullen

I've asked DSE if Vegas supports FRAME decoding, but never received an answer. DSE?

What happens if it doesn't support frame decoding but you import a frame encoded hdv clip anyway? I thought all mpeg decoders worked basically the same way and those flags don't actually do anything but tell the decoder how to output the interlaced fields.

Flags are sometimes wrongly encoded even in dvd-videos and thats why the decoders constantly analyze the picture.

You can force decoders to use weave deinterlacing (preserving the progressive nature) or the bob-method for interlaced material. Flags themselves won't stop the user from selecting the correct method (weave).

[Steve Mullen]

Originally Posted by Alex Leith

Thanks for that interesting theory Steve.

Would it still be possible to have "frame" encoding across progressive segmented frames, or are the two mutually exclusive?

That's a very good question! Since the MPEG encoding is only on the tape, it seems like only the encoder and decoder would have to deal with FRAME verses ODD and EVEN data.

Before the encoder and after encoding -- the video is field-based.

But, as I said, I'm speculating.

What I like about the DNR possibility is that so many have commented about noise as well as oil paint. The ants are really weird since they are an NTSC composite artifact. I've never seen them in HD. But, they could occur when two slightly miss-matched fields are combined into a frame.

But I don't see why they would move?

-------------

Encoders don't do the right hing without flags -- they do whatever their default action is. This may be correct, but it may be very wrong! Goggle CUE and ICE. You'll learn a lot about MPEG-2 artifacts that come from many, many bad encoders.

[Steve Mullen]

Originally Posted by Alex Leith

As I understand, it's a noise reduction technique where a pixel of one colour surrounded by pixels of a similar (but different) colour is "cored" out and replaced by the second colour. It reduces grain noise and can be used to smooth skin or sky detail. If it's used too heavily then it loses fine detail. Usually there is a threshold set so it won't smooth colours that are too dissimilar.

Looking at the image of the church and the other comparative images that Brett posed, it would appear that the paint effect happens in areas where there are similar tonal ranges.

It also looks like this effect extends to the edges of the compression macroblocks, as there are evident (16 x 16?) blocks of colour, where once there was fine detail.

This does sound reasonable too. But, why would P be cored more than I?