[M4IF Technotes] Chroma components in inter4v mode

Christoph Lampert chl math.uni-bonn.de
Mon Jun 24 13:41:58 EDT 2002


> +> Why was it chosen like that? It sound like that's exactly the wrong
> +> position, because that might be wrong for _all_ of the four blocks. 
> +> Also, it makes motion search much more difficult.
> 
> Most people don't use chroma at all in the motion search.
> (Although they would get better quality if they did, given
> sufficient processing power.)

Of course, but in 16x16 mode, a good match for lumi is most
likely also a good match for 8x8 chroma. 
In 4MV mode, this is simply not true, unless all 8x8 vectors
are very close together. One "better" lumi 8x8 vector changes
two 8x8 chroma vectors, maybe to the worse. 
Unfortunately, the value of chroma depends on all 4 lumi 
vectors, so it's not possible to search those four independently 
if chroma is to be considered :-(
> +> If would have been logical to either add a fifth vector, or 
> +> 2 bits to 
> +> signalize which of the blocks' vector to use, or something similar. 
> 
> That might not be worth the extra bits it would cost to send the
> vector or indication.  And it would take extra processing power to
> determine the values to put in those extra bits.

Are there any published results on chroma+4MV search? Or even 
only 4MV search? Any papers? It seems everybody agrees on 
"chroma is not important" without anyone really checking.
> Chroma doesn't usually consume very many bits overall
> relative to luma and isn't as visually important as
> luma, so often it is treated as not being very important
> in a codec design.  Chroma tends to take about 10-20% of
> the total bit rate, so even if you could cut that in half
> with some better processing method you wouldn't save that
> much in the overall bit rate.

So the answer is: "It was chosen that way, because it does 
not matter very much". Okay, that's an answer. 
How's the situation in H26.L? 
Christoph
-- 
Christoph H. Lampert chl   math.uni-bonn.de | Diese Signature wurde maschi-     
Beringstr. 6, Raum 14 Tel. (0228) 73-2948 | nell erstellt und bedarf
Sprechstunden: keine, aber meistens da    | keiner Unterschrift. AZ 27B-6 


More information about the Mp4-tech mailing list