Image Processing Reference
Of these, the most difficult to keep synchronized is the color because any change in
the phase of the color sub-carrier frequency will alter the hue of the color being displayed.
Problems occur if the synchronization of the color signal is compromised. Extra long
paths for the analog signal cause phasing problems with the color information due to the
propagation time. Color information runs at about 4 MHz. Wave shapes cycle every 250
nanoseconds. A cable run from one studio to another that is 100 yards long will introduce
an appreciable delay. This problem is much reduced now that the studio is a predomi-
nantly digital domain.
When sampling analog video or outputting analog video from a digital source, square pix-
els are rarely used. The actual pixel map dimensions vary according to where in the view-
ing pipeline the raster is being extracted and the kind of signal being sampled. The physical
raster is part of a transport mechanism while the image area is only a portion of that area.
Bandwidth available for recording devices such as VHS tape decks in a domestic
environment limits the number of meaningful samples, and the way that color is encoded
also affects the optimum sample rates.
The number of samples in the horizontal axis determines the fundamental data rate.
The shape of the pixels must be taken into account for some applications.
In the U.S. TV standard with 525 scanning lines, only 480 of them contain picture
information. Each line has 640 samples. This is commonly referred to as NTSC, but that
merely describes the encoding of the color information.
European TV has 625 scanning lines but only 576 of these are active picture lines.
While these will also be sampled 720 times for each line, that image must be stretched
using interpolation to 768
579 so as to achieve square pixels. The 640
480 raster must
be stretched to 720
480 to achieve the same effect. Square pixels are desirable because
they make processing of any effects or overlays more straightforward. Computation of
line drawing and circle algorithms for mattes and overlays work best with pixels that are
spaced evenly in both axes.
The values quoted here are nominal, and this situation is actually far more complex
depending on sample rates and whether component or composite source material is being
digitized. Sample rates vary to preserve color information. This relates to the frequency at
which color detail is modulated. The same Nyquist sampling constraints apply to color infor-
mation. Sufficient samples must be taken to avoid aliasing and quantization artifacts.
Nyquist states that digital sample rates must be at least twice the frequency of the highest fre-
quency being sampled to avoid loss of information. Since color sub-carriers are modulated
around the 4 MHz frequency range, sample rates in the region of 12 to 15 MHz dictate the
spacing of samples when we want to record and reproduce the color sub-carrier frequency.
This complexity is dealt with by the input circuits of the encoding system or video
cards and drivers. Most users need not be conversant with the fine points. Table 5-6 sum-
marizes the values.
The most likely sampling will be based on the CIR Rec 601 component 4:2:2 for-
mat that yields a 720
480 or 720
576 image. Your working area would be 768