[opendtv] Re: [Fwd: Re: Re: Math of oversampling - corrected links

  • From: Craig Birkmaier <craig@xxxxxxxxx>
  • To: opendtv@xxxxxxxxxxxxx
  • Date: Sun, 1 May 2005 09:57:49 -0400

At 5:14 PM -0700 4/30/05, Ron Economos wrote:
>Perhaps interlaced chroma upsampling in a progressive image?

Nothing so weird.

Welcome to the realities of video sampling and the DCT-based JPEG 
compression used to squeeze these files for distribution.

Remember my discussion with Mark about sampling theory and the need 
to properly filter high contrast transitions? Even progressive 
systems must obey sampling theory.

Diagonal lines cannot be represented in a raster of any resolution 
with high contrast transitions - you get stair steps rather than 
smooth diagonal lines. This has always be a "criticism" of unfiltered 
graphics generated for the non-Nyquist limited world of computer 
graphics. If you look closely at the image, as Doug suggests, you can 
see the same structure on the midtone grey line behind the pictures 
with white borders; but it is far less apparent here because of the 
low contrast transition from blue to grey versus the high contrast 
transition from black or red to white, then blue.

I strongly suspect that the culprit that caused the color bleeding 
(blue and red) into the white border is our old friend the DCT 
transform. When we quantize the DCT we are making adjacent samples 
more alike; one of the artifacts of this is to change the values of 
the samples, and color bleeding is very common. You can also see the 
artifacts of the DCT transform around the edges of the numbers in the 
grey area below the pictures with white borders. This is the classic 
"mosquito noise" that occurs when you have samples with high contrast 
inside a single DCT block.

You can UNSUBSCRIBE from the OpenDTV list in two ways:

- Using the UNSUBSCRIBE command in your user configuration settings at 

- By sending a message to: opendtv-request@xxxxxxxxxxxxx with the word 
unsubscribe in the subject line.

Other related posts: