This paper is composed of 6 sections including intro-
duction and conclusion. Section 2 outlines the Fourier de-
scriptors of closed and open lines. Section 3 describes the
centroid-based transformation in the spatial and frequency
domains. An algorithm of least-squares matching in fre-
quency domain and interpretation of the results from the
matching algorithm are illustrated in section 4. Section 5
presents some experimental results using synthetic data.
2. FOURIER DESCRIPTORS
2.1 Closed lines
A two-dimensional closed line can be described by two
periodic functions z(t) and y(t) (Fig. 1). The parameter t is
defined as 2l/ L, where L is the perimeter of the closed line
and | denotes the arc length along the line from the starting
point s to p. According to the theory of elliptic Fourier de-
scriptors [Kuhl and Giardian, 1982; Lin and Hwang, 1987],
these two periodic functions can be expressed by Fourier
expansions in matrix form as
9] - [s] E[R £] [E]. ©
k=1
= 7 elt)dti o=#Fu0d
ar = i fy a(t)cosktdt; b, — 1i f" z(t)sin kt dt;
cy, — 1 fj" y(t)cosktdt; d, — ijj" y(t)sin kt dt.
In Eq.(1), ao and co are the mean values of z(t) and
y(t) respectively, which indicate the geometric center of the
closed line, or so called the centroid.
(t) y(t)
| =
0 2m et Ex et
Fig. 1. A 2-D closed line and its periodic functions.
2.2 Open lines
An open line is traced once and then retraced back-
ward so that a closed boundary is obtained (Fig. 2). The
Fourier descriptors can then be applied. Let L denote the
arc length of an open line and the parameter t is defined
as wl/L. The functions of z(t) and y(t) can be expressed
470
as periodic functions. A close examination of the peri-
odic functions (Fig. 2) yields two important characteristics.
First, they are even functions because z(—t) = z(t) and
y(—t) = y(t). This implies that the coefficients of b, and
d, are all zeros. Second, the integration f; z(t) cos kt dt is
equal to that of £r æ(t) cos kt dt, and it is appropriate
to y(t) also. Therefore, an open line can be described with
the Fourier expansions as
EEE], a
where
ao = t fs (t) dt; co = = Jo y(t) dt;
ar = $ fy z(t)cos ktdt; cy — 2 Jy y(t)cos kt dt.
(t)
|
MM
M.
di T -
2n-t2 2n-t# O7,
Fig. 2. A 2-D open line and periodic function of z(t).
3. CENTROID-BASED TRANSFORMATION
AND PHASE SHIFT
3.1 Transformation in Spatial Domain
If a linear feature consists of a list of (z, y) coordinate
pairs of nodes, a transformation in spatial domain is im-
plemented by transforming all coordinate pairs in the list.
Conventionally, such transformation is operated about the
origin of the coordinate system. For instance, let the list
of (z', y') be the coordinate pairs after transformation. A
similarity transformation about the origin is expressed as
z' cos —sin0 z Az
MES msi ol, (3)
where
S Scale factor;
0 Rotation angle;
Az, Ay Translation.
With this transformation, one can easily discover that
the positional change of the transformed feature does not
correspond with the translation parameters Az and Ay, be-
cause the centroid of the feature is changed by scaling and
roi
of
fez
m:
iso
pli
irc
exi
pre
wh
pai
par
the
she
can
anc
be
Let
tra
Zer
trix
ma
sep
irai
nat;
ficie
the
whe
3.3
of (