Digital Signal Processing Reference
In-Depth Information
Mosquito noise tends to occur around text or letters in an
image, or with computer-generated video objects with very sharp
transitions. Mosquito noise looks like a cloud around the edges of
text and computer generated graphics.
Reducing this artifact requires spatial image analysis. Each
region, or even each pixel, must be classified into categories such
as edge, texture, flat and artifact regions. Across the temporal
dimension, or sequential video frames, motion prediction can
also be used to help distinguish moving edges due to objects in
motion. A combination of this analysis is used to identify artifact
regions and apply filtering techniques to reduce in those regions.
This tends to be an adaptive process.
Sometimes the only choice may be to accept the noise, or to
filter and have blurring occur. When motion is very rapid, filtering
may be the better choice, even if blurring is introduced, as the
blurring is often not as obvious in fast motion scenes. And on
some occasions, the motion is too rapid for the temporal rate
(frame per second), violating the sampling theorem, so that some
form of degradation due to the aliasing is inevitable.
16.3 Block Artifacts
Block artifacts are a discernable visual blocking effect on the
video which can occur at the 8
8 boundaries used in the DCT.
This is due to several effects, one of which is differing DC coeffi-
cients. Insomecases, this effect canbecomevisuallyobjectionable,
as shown on the left side of
Figure 16.5.
Each block can also have
varying amounts of quantization, depending upon how many
AC coefficients have significant amplitude. When high quantiza-
tion is used, the DCT averages each 8
8 block approximately,
making it roughly appear like a group of pixels, each of size 8
8.
These effects can become even more pronounced in scenes with
high motion, such as sports broadcasts.
A common solution is to filter at the block edges, to smooth
the transition between blocks. This filter must by necessity be of
short order, and apply to only one or two pixels at the 8
8
boundary. This filter is adaptively applied and is part of the H.264
decoder in some profiles. One of the factors driving it is the
quantization level: at high quantization levels, blocking artifacts
are more predominant, and deblocking filtering is needed.
Another factor is video content: a smooth region will have
a stronger blocking effect compared to an area with a high-
activity video level. The smoother region will require more
aggressive deblocking filtering.
