resizing AFTER compression

[frankio]

Would it be possible to implement a such thing for DVD conversion purpose?

I mean putting resizing algo after compression algo in any MPEG-4 codec, in that way the codec would compress better having more information in a full-screen frame than in a resized one.
The codec dialog should you permit which resolution to choose.

Thus we can use just one stats for every desired output resolution.

What do you think of?
I don't know if it is a good idea

[gldblade]

>I mean putting resizing algo after compression algo in any MPEG-4 codec

There is no resizing algo in MPEG4 codecs. That's handled by the program you are using (i.e. VirtualDub, Flask).

And you can't resize the video stream after it's compressed unless you do it on playback.

[duartix]

Purists will tell you that the stats file will depend a lot on the resolution and also the resizing algo (bilinear, bicubic, precise bicubic etc) but I believe that in most cases it will vary linearly so you can use the same stats file without too much a diference.

Unless you have a too much textured source however and resize to a lower resolution you can loose the texture and have a stats file that will give values not near to the truth, you will be okay...

...thats just my opinion, I could be wrong...

[xzquala]

in order to resize the video, it would have to be decompressed. so you would still have to resize it again. the only thing allowing you to resize on playback is the fact that the resize step is done after decompression

[frankio]

Sorry, I didn't know resizing requires uncompressed frames.

Anyhow, isn't there a sort of pipeline in a video encoder?

1. pre-precess (gather maximum information about full-res frame-->stats)
2. resize frame
3. compressed delta-frame (DCT?)

Just a try

[frankio]

It could be the first step for a STANDARD stats file (common with DivX3, DivX4, Xvid etc.), thus allowing everyone to have his favourite output resolution from the same stats.

All done feeding maximum info to Motion Prediction algo and every pre-processing stage.

Maybe I speak a wrong language?