‚4250
6 WHY FEEDBACK MATCHING AND BANDS?
1} To
2) To
3) To
4) To
facilitate the coarse-to-fine iteration process.
promote the spread of correlation.
locate automatically all homologous pixels in the window.
eliminate automatically detail present only in one image.
Stereo correlation involves multiplication of input data followed by integra-
tion or the summing of products over an area. The larger the area that can be
integrated the less the noise or roughness in the Z signal. We define the rel-
ative coherence area as the largest circular portion of the stereo image areas
that can be regarded as identical and over which dx is constant. Integrating
beyond the limits of the coherence area sharply increases Z noise and reduces
the reliability of the Z data.
A major problem in the design of stereo correlators is the small coherence area
of stereo imagery before transformation, typically about one video cycle in
both X and Y. Therefore the low frequency video components, representing large
image detail, have a larger
coherence area than higher
frequency video components
representing small image
detail.
Dividing the video spectrum
into frequency bands allows
the integrating area to be
separately matched to the
coherence area for each band
and also provides a basis for
progressive feedback matching.
Figure 2 illustrates the pro-
gressive clearance of disparity
for an object in the form of a
layered pyramid. A large B/H
ratio of 1/1 has been used to
exaggerate the differences
between the stereo images. The
cameras are placed at infinity
to simplify the drawing.
Unmatched parts of the images
are shown diagonally hatched.
Matched parts are shown clear
and parts seen by only one i
camera are shown black. Four
band iterations are shown and
are sufficient to illustrate
correlative matching on the
simple object chosen. Six or
seven bands are required to
resolve fully the imagery of a
typical 500 by 500 pixel patch.
OBJECT
STEREO
IMAGES
AFTER BAND 4
TRANSFORMAT ION
AFTER BAND 3
TRANSFORMATION
AFTER BAND 2
TRANSFORMATION
AFTER BAND |
TRANSFORMATION
1072
Figure 2
PROGRESSIVE: DISPARITY ‘CLEARANCE