Digital Signal Processing Reference
In-Depth Information
a
b
c
Fig. 27
2D non-separable filter: (
a
) ideal, (
b
) impulse response and (
c
) magnitude response:
contour plot for
−
3dB(
inner curve
),
−
6dB(
middle
)and
−
30 dB (
outer curve
)
4.2.2
Passband Approximation with a Separable Filter
The computational complexity of a 2D filter is rather high and considerable
computational savings can be achieved when it is possible to separate the 2D filter
into two 1D filters, one filtering in the horizontal direction and one in the vertical
direction. One approach to the design of separable passband-optimizing filter is
Based on the known interdigitation pattern and angular visibility of each element,
one can derive the pattern of visible pixels as seen from the sweet spot of one view.
a source of aliasing. In order to avoid aliasing (Moire artifacts), a filter has to be
designed in such a way that its passband does not overlap with any of its copies
generated by moving its center to any of those aliasing sources. It is possible that
there are several different separable filters that can be used as antialiasing filters
antialiasing, but due to different shapes the visual quality of displayed images will
be different. Which separable filter would yield best visual results depends on the
as subtitles, filters with a passband close to square perform better than filters with
elongated passbands. For designing 1D filters with the desired cut-off frequencies,
the windowing technique with the Kaiser window of length 24 can be used (see also
6 dB contour) of several
−
4.2.3
Passband Approximation with a Tunable Filter
always give the best perceptual quality. Some people prefer sharper-looking images
at the expense of some Moire artifacts. In order to allow the user to control the