Cutting Out the Noise - A Practical Guide to Video and Audio Compression - page 582

the temporal domain, that is, at a single X-Y point but also from one frame to the next. This
is analogous to grain removal when processing films, and the same grain-removal filters
might help if you have them.
Removing the noise within a single frame is a compromise but achieves nearly as
good a result. It might cause some average-intensity changes from frame to frame but
these will be small.
De-noising an image effectively is computationally very expensive. A reduced CPU
loading is achieved by applying a high-frequency cutoff filter that operates only in the
horizontal axis since this corresponds to the transmitted line of video, which is what
the noise is being imposed on. The cutoff might destroy some edges that you want to pre-
serve.

36.3 Different Kinds of Noise

Certain kinds of noise, such as dust and scratches, should have been removed earlier in
the process. Some filters are introduced to remove ringing artifacts and also to average
pixel values between frames so as to remove the inherent white noise that is present in
analog transducers (cameras).
Analog noise cannot easily be removed because the effects extend over an area of
the picture, and decisions about where edges of the noise pattern are situated are difficult
to build into an algorithmic framework. The human eye does a remarkable job of
discriminating between noise and genuine picture information. If you concentrate you
can observe the noise, but after a while you tend to tune it out and ignore it. While it's
good that the eye and brain work cleverly together like this, it is rather tiring for the
viewer and it's better to remove the noise if you can. Your video compressor will do a
better job, too.
Repeating and cyclic noise artifacts can be removed algorithmically if you can iden-
tify the cycle and pattern. The effects of signal interruption and interference from other
radio-frequency sources generate these cyclic effects. If you live near an airport, for exam-
ple, you may notice a repeating pattern of dashes that scrolls down across the picture peri-
odically. These are the radar pulses that interfere with your TV service reception, causing
momentary loss of signal.
For the amateur or semi-professional who may actually be working in a domestic
environment, there are a lot of noise sources to cope with. You may be able to alleviate
some of these effects by making sure your systems are running on a clean mains supply,
so introducing mains filters could be worthwhile.
Electrical noise from appliances, car engines, lawn mowers, electric shavers, and cen-
tral-heating systems all contributes impulse noise that can be removed by careful applica-
tion of dropout compensation. You might take a few pixels from a previous or subsequent
frame and paint in the missing detail.
There are also various herringbones and dot-crawling artifacts that might be intro-
duced by the analog production processes on older footage.