You roughly have 4x as many pixels with 8 bit depth, when transfer HD to SD. So have roughly 4 HD pixels per 1 SD pixel. Can gain additional depth for each SD pixel by averaging the 4 pixels, making roughly 10 bit 4:2:2 MPEG4 SD.

Is there program to do conversion, like FCP, maybe? Is there PC program?

After your editing, color corrections in 10 bit, in some kind Digital Beta quality, except it's MPEG2, can transfer to high quality SD. Or is better just edit in HD(V), then do transfer? If is the case, at least is explanation, why some say that when transfering FX/Z1 to SD, get better quality compared to SD originating footage, that get near Digital Beta quality.

Radek

Radek I have been working on a program to convert HDV 1080 video into RGB 4:4:4 10 bit video.

Thomas,

That is interesting. Once you remove MPEG2, have so much information to work with. On some scenes frames become 4 Mb large. At 30 fps it is 120 Mbps. That is close to HDCAM. On fast moving scenes frames can be small as 1.2 Mb. That's 36 Mbps.

Radek

Radek I have been working on a program to convert HDV 1080 video into RGB 4:4:4 10 bit video.


Thomas, I assume you mean that you are coding a program that will convert HDV 1080 to 10 bit 4:4:4 RGB. Or, do you mean that there is a program out there already and you are learning how to use it?

Either way, please keep us posted. I would be especially interested in learning more about this if you are coding from scratch. Are you a programmer?

Yes I am making it.

Do you have any more details on your project?

I am working on a tool for PC and MAC to convert 8 bit 1080 HDV footage into 10 bit RGB 4:4:4 SD footage or at least as close as I can get. This tool will also remap video frames and audio from 25 down to a perfect 24p.

I am still running some image processing tests but I am getting very close to making it work.

Good luck with it. When you have a working beta, please bump this thread. I'd love to check it out.

I think if you do editing in HD(V), will retain this fine information and if transfering to 8 bit SD, it will be like transfering from 10 bit SD.

Radek

Hello.


Could you explain how you are going to aquire a 10 bit SD image out of a HDV footage which is just 8 bit? As far as I can imagine there is no way to gain any additional bit depth out of something that simply has not it.

Am I wrong?

Radek got somewhat close with his first post. Well kind of at least. There is a little bit more to it but it was the right direction. I cannot say more than that about the 10bits right now.

as for the 4:2:0 to 4:4:4 well that is a pain in the but.

4:2:0 in my opinion is a evil form of compression. Even 4:1:1 is better. At first glance 4:2:0 can seem to be better because the chroma pixels are spaced out a little bit better. The problem though is that every single chroma pixel is interpolated. With YUV with Y being luma and U and V being the two chroma channels the U and V never matches up.

4:2:0

YU-,Y--,YU-,Y--,YU-
Y-V,Y--,Y-V,Y--,Y-V

Because of interlaced video the chroma is shifted in between every other line so you actually end up with something like.

YU-,Y--,YU-,Y--,YU-
YU-,Y--,YU-,Y--,YU-
Y-V,Y--,Y-V,Y--,Y-V
Y-V,Y--,Y-V,Y--,Y-V

This means the U samples are accurate for line 1 but duplicated for line 2,3 and 4.

your V samples for line 1 and 2 are interpolated. That means every single pixel in your image has either an interpolated U or V pixel. You never can have pixel accuracy.

If we sample 25%, 720x540 from 1440x1080 we will now get,

YU-,YU-,YU-,YU-,
Y-V,Y-V,Y-V,Y-V,
YU-,Y-U,Y-U,Y-U,
Y-V,Y-V,Y-V,Y-V,

We now have a 4:4:? I still don't know what to call what we end up with. It is kind of a 4:4:2 except each chroma pixel only contains accurate U or V data. On top of that the U and V samples are duplicated and interpolated from line 1 to 2 and so forth. You would almost need to blend the data from line 1 and 2 to try to get at least a close interpolation of U and V samples together.

There are other ways of trying to interpolate and smooth out the chroma but I am shooting for the most accurate pixel data that I can.

I know Graeme is working on a 4:2:0 upsampler. Maybe he is having better luck I am.

Does all this look better even with my complaining? It sure does. It does look much better but it just isn't perfect.

If mpeg2 would have used 4:1:1 at least I could have given you a real 4:2:2

YUV,Y--,Y--,Y--,YUV
YUV,Y--,Y--,Y--,YUV

to

YUY,Y--,YUV,Y--,YUV
YUV,Y--,YUV,Y--,YUV

Every other pixel is interpolated but at least the pixel before it has an accurate U and V sample.

I am testing out a few new methods to see if I can come up with something nice.