[opendtv] Re: Precision



Craig Birkmaier wrote:
> Good points. As I just noted in a response to Tom, excessive
> quantization can cause major problems in the areas of high frequency
> edges. We can only provide accurate edges IF we do not quantize them
> excessively. Unfortunately just the opposite is what is happening. The
> edges are the first thing to fall apart when we compress too much,
> whereas, in blocks with uniformly distributed fine details we just
> replace those details with correlated noise.
>
However with oversampling the high frequency edges are filtered down to Nyquist of the target resolution. Theoretically I think that means lines and edges can move smoothly in pans (relative to maybe 3 pic heights viewing distance, invisible raster) as long as the higher (but still < Nyquist) frequencies are not lost in compression.

Though I have concerns the Nyquist frequency for 8x8 DCT blocks may be a bit less (7/8 or less?) than would be predicted by the video resolution. I'm not sure about this one.

But not that, however the human visual system interpolates, I believe our perception of where an edge is located is likely a bit more precise than the pixel grid as long as Nyquist is not violated, even with slow pans. (possibly unsupported conjecture)

- Tom

At 8:16 PM -0400 5/28/07, Allen Le Roy Limberg wrote:

The discussion seems to dwell on the type of resolution sometimes referred to as "fine line" resolution -- i.e. how small a detail can be reproduced. An important type of resolution in displays is "edge" resolution -- i.e. how accurately a diagonal line can be positioned and how well well "jaggies" can be suppressed. This doesn't involve extended frequency response, but rather accurate phase response of the frequency components of the video image. Jim
Davis out at RCA avionics in Van Nuys, CA pointed this out in a patent I
filed for him 35 years ago or so, as I recall.

Fewer lines in the display hurts the angular resolution of an edge and
lowers the step frequency making jaggies more visible.  The human visual
system is quite sensitive to jaggies. We easily espy low-spatial-frequency parts of a display being out of correct spatial or temporal position, while
superposed random noise is largely disregarded.

Video compression may have fewer problems with accurately defining phase
response of the moving edge than with defining moving fine line detail.



Good points. As I just noted in a response to Tom, excessive quantization can cause major problems in the areas of high frequency edges. We can only provide accurate edges IF we do not quantize them excessively. Unfortunately just the opposite is what is happening. The edges are the first thing to fall apart when we compress too much, whereas, in blocks with uniformly distributed fine details we just replace those details with correlated noise.

Regards
Craig


----------------------------------------------------------------------
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.



--
Tom Barry                  trbarry@xxxxxxxxxxx  



----------------------------------------------------------------------
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: