Image Processing Reference
In-Depth Information
(a)
(c)
(e)
(b)
(d)
Fig. 11.5. Electronic circuit and transistor detection.
a,b.
the original and the prototype.
c,d.
|I
20
|
and
I
11
.
e.
GHT accumulator with 2
π/
256 angle resolution (from [22]). The
circle
is
inserted to show the location of the peak
A
(
x
j
,y
j
)=
l
δ
(
θ
(
x
j
+
x
l
,y
j
+
y
l
)
−
φ
(
x
l
,y
l
))
(11.70)
where
l
runs over the edge elements of the prototype, and
φ
(
x
l
,y
l
) is the direction
of the prototype edge. The
θ
(
x
j
+
x
l
,y
j
+
y
l
) corresponds to the direction of the
edge at the position (
x
j
+
x
l
,y
j
+
y
l
) in the contour image, whenever there is an
edge pixel, or
i
is purely symbolic and serves to make
θ
different from
φ
, because expected contours modeled by the prototype may be
missing in the contour image. In such instances this will generate a zero contribution
(vote) from the
δ
-function. The array
A
(
x
j
,y
j
) represents the vote accumulator and
can be constructed on the same grid as the image itself. Figure 11.5 shows the result
of detection of a transistor in an electronic circuit picture.
∞
. That
θ
is defined to be
i
∞
11.6 Voting in GST and GHT
The parameter estimation of the generalized structure tensor is more than a correla-
tion of edge magnitudes. It carefully takes into account the directions of the edges,
too. Here we show that the GST detection extends GHT, to a complex GHT in which
the votes are allowed to assume complex values.
The GHT accumulator,
A
(
x
j
,y
j
), can be compared to an estimation of
I
20
at
local images around all image points,