Game Development Reference
Figure 13.13 Base images for different filterings (these images can be viewed in 3D as in figure 13.1)
Figure 13.14 Adaptive blurring: where the disparity and local frequency are incompatible and where
high frequencies were suppressed
by wavelet processing and is then reconstructed on the basis of wavelet coefficients,
except those corresponding to high frequencies.
This method of processing is very effective but too slow to be applied in real
time. And yet, we have seen that to avoid a pseudoscopic movement, it is important to
generate images in real time. A simple method for resolving this problem is to carry out
this processing by taking advantage of the calculation capacities of the graphics card
of the computer. It is possible to apply a Gaussian blur or use a post-processing moving
average (or directly on the textures of objects). The intensity of this blur depends on the
disparity of the processed point, always according to the comfort function of Wöpking.
18.104.22.168 Experimental results
Jérôme Perrin tested the effect of his processing on visual comfort with fixed stereo-
scopic image pairs. The testers had to move a ring around a wire without touching it.
According to the tests, the images were either filtered or not. Each experience lasted
for some time so that visual strain could be felt. Subjective comfort tests showed that
adaptive blur filtering is definitely better than no filtering.