Signal Processing for Stereoscopic and Multi-View 3D Displays - Signal Processing Systems - page 37

In signal processing, imaging is tackled by an anti-imaging post-filter. In
spatially-multiplexed displays, imaging is created by the physical structure of the
display; it is therefore impossible to impose a post-filter. However, the visual
perception of imaging can be partially mitigated by a pre-filter. Consequently this
filter can be merged with the anti-aliasing pre-filter. In order to determine the
properties of the required combined (anti-aliasing and “anti-imaging”) 2D filter it is
necessary to determine the performance of the display in the frequency domain; that
is, we have to know which frequency components in the image we can keep (ones
that will be properly represented on the screen), and which ones we have to attenuate
as potential causes of distortions. A proper design of the filter should result in the
best possible representation of images on the display, minimizing aliasing, imaging
and ghosting. It is worth mentioning once more that it is impossible to remove
imaging artifacts since they are caused by the display's optical layer. However, some
of them can be reduced to a level at which they are less disturbing.
A measurement-based method for deriving the frequency response of the display
(display passband) is described in Sect. 3.2.4 . In the text, the region containing
frequencies that are properly represented on the screen is denoted as passband ,and
all other regions as stopband. In order to improve the image quality one should
design a filter which attenuates frequency components in the stopband. The methods
in the following two sections present an example approach for designing such filter.

4.2.1 Passband Approximation with a Non-Separable Filter

The design of non-separable passband-optimizing filter is discussed in [ 69 ] and[ 70 ] .
As a practical example, such a filter is designed for a 24-view 3D display which has
a passband as the one in Fig. 20 , bottom-left. For that display the shape of the ideal
2D antialiasing filter is as shown in Fig. 27 a . In this figure the curve shows the ideal
cut-off frequency; that is the passband of the filter should be inside the contour and
its stopband everywhere else. For designing a non-separable 2D filter approximating
this ideal one, the windowing design technique with the Kaiser window of length 24
has been used [ 79 ] . The Kaiser window has been selected as a good candidate due to
its relatively narrow transition band and flexible attenuation. The variable parameter
of the Kaiser window controlling the stopband attenuation has been set to

2.2.
Such selection ensures a stopband attenuation of at least 30 dB that is good enough
for the display under consideration. A filter size of 24 by 24 has been chosen as
a good compromise between the implementation complexity, transition bandwidth
and approximation of the ideal filter.
The design results in the 24 by 24 2D non-separable filter with an impulse
response, as shown in Fig. 27 b . The corresponding magnitude response (contour)
of the designed filter is shown in Fig. 27 c . The

β =

6 dB line in Fig. 27 c approximates
the ideal cut-off frequency. Due to the finite transition bandwidth of the designed
filter, even after applying it to the input image, some aliasing errors will occur on
the display. However, the aliased frequencies will be attenuated by the filter (either
filter transition band or stopband) and as such they will not be visually disturbing.

−