[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [video] about DCT




> Hi Gents;
> I have some questions in my mind, may be one of you might know it.
> First one; Is it possible to process RGB data which was gathered from a
> VGA signal into a DCT module such as Richard Herveille's DCT algorithm?

Yes, no problem. The DCT doesn't care what kind of pixel information you feed 
through it.

> Of course before going into the DCT process, there will be a RGB to
> YCbCr conversion..

If you want to. It is not necesarry. However if you use it to do a 4:2:2 
conversion too, than your required bandwidth and computation power reduces 
(see below).

>
> Secondly; I have been thinking on Richard Herveille's valuable work on
> DCT. It accepts 8 bits of video data. I think that is for first row of
> pixels (8pixels each line).

Correct, it computes on a matrix of 8pixels * 8lines.

> But as you might realize that after RGB2YCbCr
> conversion each pixel will have a ~8bit of information whereas DCT module
> accepts only 8 bits of information at each clock cycle. So is there
> something wrong the way I think about? Or it just accepts 8 bits of
> information for each pixel at a clock cycle. I am confused about it. Could
> anyone kindly explain the process?

The DCT algorithm computes on a 8x8 pixel block. However for multicolor pixels 
you need to do a multicolor DCT. This means that the DCT algorithm operates 
on one color only. Therefore for RGB you need to computer all three colors 
per pixel. This means that for an RGB pixel you need to do 3x as much 
computations than for a b/w pixel.
This is where RGB(4:4:4) to YCrCb(4:2:2) provides a significant improvement. 
You only need to do half of the computations for the color components, 
thereby reducing your computation requirements considerably.


> Or is there any example source code that
> I can look into from begining of buffering of video data and sending them
> into a DCT module?
> Or someone kindly tell us about the process in general? I have looked at
> the source code and some other references about DCT on the net but
> could not find an appropriate information.
>
> Cheers,
> Aris

--
To unsubscribe from video mailing list please visit http://www.opencores.org/mailinglists.shtml