[opendtv] Re: Math of oversampling
- From: "Don Munsil" <don@xxxxxxxxxx>
- To: <opendtv@xxxxxxxxxxxxx>
- Date: Fri, 22 Apr 2005 11:10:15 -0700
From: "Bob Miller" <bob@xxxxxxxxxx>
> If a broadcaster uses a 1080p camera and broadcast as 480P and on the
> reception end it is upconverted to 720P or stays 480P I understand that
> because of oversampling both the 720P or 480P image would be better than
> otherwise would be expected. What I would like to do is quantify this
> value. How would you compare a straight thru 720P broadcast to one such
> as that above. Would the 1080P>480P>720P route be 75% of the quality of
> the 720P>720P>720P route? Would 1080P>480P>480P be 110% of 480P>480P>480P?
It's not something that's simply quantifiable. A very clean 480p image
gathered with an excellent 480p camera or telecine could be every bit as
good as a 1080p image scaled to 480p. Practically speaking, the 1080p->480p
signal will often look better, largely because of preservation of edge
sharpness. An image from a 1080p camera downconverted to 480p by a
good-quality algorithm like bicubic will probably have better inherent edge
sharpness than a raw unsharpened feed of a signal from a 480p camera.
On the other hand, judicious application of quality sharpening algorithms to
a native 480p image can quite possibly get it to very nearly the 1080p->480p
image. It really depends on the cameras, the algorithms, the various formats
the picture goes through, etc.
> I also have been made to understand that the bit cost of a 480i
> broadcast, say X, would not be 2X that of a 480P signal. More like 1.4X
> but with the extra juice added by the 1080P acquisition maybe more like
> 1.5 or 1.6X.
I assume you mean the opposite - you would expect 480p60 to be 2X the cost
of 480i60 (480i30 if you prefer Poynton's nomenclature). Yes, it's not
actually 2X the bit rate requirements. The requirements, however, change
depending on whether you're talking about deinterlaced 480i60 or native
480p60.
Deinterlacing 480i60 to 480p60 doubles the number of MPEG pictures
transferred, but the efficiency of both the DCT and the motion estimation
goes up because deinterlaced video has higher interframe correlation and/or
lower vertical resolution, and I would say somwhere between 1.2-1.5x the bit
rate for similar quality is about right, depending on the material. However,
1080p60 or 720p60 scaled to 480p60 produces 60 completely independent frames
per second, and I would be surprised if you could compress true 480p60 to
the same quality with less than 1.6x the bit rate of 480i60, and at times it
would require rates closer to 2X. If I had to guess, I'd say 1.6x-1.8x.
However, interlaced encoding severely compromises the overall picture
quality, especially because of the deficiencies of 4:2:0 interlaced chroma
encoding, so perhaps you could pull back the bit rate somewhat and rely on
the fundamental visual improvements of progressive encoding to produce a net
gain in perceived quality.
I would very much guess that most people would be very happy with the
quality of true 480p downconverted from 1080p, 720p, or even 1080i. Most of
the 480p the public has seen has been deinterlaced 480i (like the 480p our
local Fox affiliate was broadcasting last year). Real 480p60 can look
astonishingly good.
Don
----------------------------------------------------------------------
You can UNSUBSCRIBE from the OpenDTV list in two ways:
- Using the UNSUBSCRIBE command in your user configuration settings at
FreeLists.org
- By sending a message to: opendtv-request@xxxxxxxxxxxxx with the word
unsubscribe in the subject line.
Other related posts:
