I loved the clips of Kaku's XL-H1. The night shots on the highway are very impressive.

HOWEVER the street lights and the car lights are very blocky. I don't know how to post a still on this forum to make this point.

So is this an artifact of the limitation imposed by 4:2:0? So it would not be like that with 4:2:2 HDI output?

Frankly I don't see why one has to comprise in resolution for the colors. I mean the camera is 1440x1080, but for color it is only 720x540? WTF??? Why doesn't all video be in 4:4:4?

What will Blue-Ray be? 4:4:4 (ideal) 4:2:2 or crappy 4:2:0?

Thanks.

Well the motion picture industry would agree with you, which is why sony was forced to make a high-high end camera (HDCAMSR) to their high end camera range (HDCAM).

The problem is 4:4:4 video, especially at HD resolutions has way too much information to record every n/th of a second. Which is why if you do want to record uncompressed HD, you need a large, fast, heavy and expensive raid array of hard disks. They just can't get that much information on or off a tape at those speeds.

All tape formats that record HD have some kind of compression involved.

Blue-Ray will also be 4:2:0 (just like current DVDs) for the same reason.

So is this an artifact of the limitation imposed by 4:2:0? So it would not be like that with 4:2:2 HDI output?
I haven't seen that particular clip, but I would point you towards this page on Adam Wilt's site, so you can see some examples of 4:1:1 vs. 4:2:0.

http://adamwilt.com/pix-sampling.html

4:1:1 and 4:2:0 both offer one color sample for every four luminance samples. 4:2:2 offers one color sample for every two luminance samples, and 4:4:4 is of course one color for each luma.

To be honest I can't see the "blockiness" but if it is there is more probably from compression artifacts than subsampling of the color. In these compression schemes blocks of data are formed and these subject to transform processing and then motion analysis. These schemes are not lossless and where they err the visible result is often the appearance that the picture is overlain with a mosaic or bits of the picture look like tiles which don't quite fit in.

The reason 4:4:4 is not used is because it doubles the amount of information required to be transmitted or stored relative to 4:2:0 or 4:1:1 and for most people for most scenes most of the time the extra color information isn't perceivable. "Subsampling" of color has been practiced in the TV industry since the very first days of color broadcasting and continues to this day. HD would be impossible (with today's technology at reasonalble $) without it. Yes it discards information and where detailed color information is required (chroma keying) it can be a problem.

Actually, given that you're compressing the video, you can get better pictures and lower data rates by recording Y'CbCr 4:4:4, but with compressing the Cb and Cr to a much greater extent than the Y' component. By using that method, the compression will give you detail in the chroma where you need it, and let it go blocky where you can't see the blocks, giving low data rates, but a visually superior performance.

Also, you can use reconstruction techniques upon the Cb and Cr if it's 4:1:1 or 4:2:0 which give you back a lot of the detail.

I think what you're seeing is straight 4:2:0 without interpolation, which is not how a viewer would ever see it.

Graeme

Well I finally managed to upload a frame of the clip of Kaku:-

http://home.comcast.net/%7esamgold/photos/canon.jpg

See the license plate on the mini van. Nice and clear you can read "45-54" but look aht the red lights - on the cars and the buildings. All blocky.

I must be blind because...

I can't read any numbers off the plate...

Hse Kha,

Yes, this is the typical look of 4:2:0 colorspace. If this were 4:1:1, you'd see something similar, but more like horizontal bands instead of blocks. It is especially apparant in high-contrast, heavily saturated reds, greens, and blues.

Products like Magic Bullet can intelligently restore chroma detail back to 4:4:4. Some NLE's have a deartifact, but it's usually just a chroma blur, which isn't as good.

Josh

Joshua,

Thanks for confirming what I thought. That's why I hate 4:2:0. But if anything I still think it is way better than 4:1:1 (from DV) which looks even worse for the same bitrate.

Anyway it's a shame that for just 33% more bits (16 for 4:2:2: vs 12 for 4:2:0) HDTV has decided to go for 4:2:0, even the Holy Grail of HDTV - Blue Ray! Surely with H.264 at 25mbps they could have gone 4:2:2, but no.

Oh well...

I think 4:2:0 is nice in that it distributes the loss vertically and horizontally.

However, for interlaced material, I think 4:1:1 is superior, because each scanline is whole in and of itself. At least it is (in)accurate only with regard to the current scanline.

With interlaced 4:2:0, the luma channel contains two discrete points in time, and the color channels are a compressed combination of both of those moments. It seems very inaccurate to me.

I agree. I find 4:1:1 to be much better for interlaced video. 4:2:0 is a pain with interlaced video. True progressive HDV looks very good with 4:2:0 because it is a similar type of chroma compression as a jpeg image.

At the same we do have to remember that HDV is a consumer format and 4:2:0 is fine for most things. As much as we want to think of HDV as professional it just isn't. It is no different than DV really in terms of the market it is aimed at. 4:2:2 HDV would be nice but the 25Mbit rate just wouldn't handle it very well. We have people complaining now that 25Mbits isn't enough at 4:2:0 let alone try to cram 4:2:2 into it.

At the same time remember all of the DVD's you buy/rent and watch are also 4:2:0 and nobody has ever really complained about it.

Joshua, that is a good point about 4:1:1 being better for interlaced video. It makes sense. However even with the SDI 4:2:2 output, ultimately the footage will end up on HD-DVD/Blue Ray or be Broadcast, and when the happens it will wind up as 4:2:0 sadly...