AUTOMATIC OBJECT RECOGNITION
IN LINE MAPS
44 Stockholm,
method used
with
the
data
of
In
to treat the
systematically, the
formalized.
The first
The
made
1:2000
composed
buildings
on the type
The second part of this
trying to implement the
on IBM PC. The
programming the rulespossible situations and
The result gave
of second order
rules,
of
100%
paper
Turbo
the
to
input data ..
1. INTRODUCTION
A lot of efforts have been involved
automatic
of symbols. The classical way to
symbol
been training the
for the classification task
cons
and
the relevant
function for each symbol
reports results with up to 99%
handwritten
out that
letters, (Holbaek-Hanssen et al", 1986).. Weber
there are several companies which
offer
symbol
method
recognition without giving details about
that
and results (Weber, 1987). It seems
recognition of numbers, letters and
ects
long and successful tradition, while
based on vectorized data still lies on its early
De Simone reports a success rate of 90% with 1% error
automatic recognition of railways, roads and landparcels, (De
Simone, 1986).. He uses the combination of the geometrical/
statistical and the relational method for recognition, (Weber,
1987) .. The check of neighborhood relation of the
candidates decides
these are refused or accepted into the
class
question
(e . g . ,
the
a
of a
i
are
cons
the model at the
mapreading
the model as far
abil
and
come
pointed out ..
In
symbol IS
reality.
boundary marks
the
a
not related to the
of
are the
An
related to
on the
, streets and real
treated in
work refer to
2.2 Perceptual background
A map shows a simpli
mapreader
variables and
visual variables
raster (
visual si
variables 1
other hand of
contrasts,
and
ect I s
their relationships
transformed to
reality. Therefore the
to other objects (e.g.
be taken into account
classification a human
and
the
of these ..
and
are
The
the
ects
ect and
o
To
fferences
doesnWt demand
of
level
on
work
the
means
et
of
and
are
for
INFORMATION
=
~
MESSAGE +
~
RELATIONSHIP
1)
(
..
E .. g ..
a
consist
1
(
where
An
the
can be written as:
L ELEMENTARY MESSAGE
for
L RELATIONSHIP
lustrated
in
level
level
n-l
level 1
ELEMENTARY
Both the
different
and I
ELEMENTARY
MESSAGE
ELEMENTARY
MESSAGE
and the
a I
can have values I
can have
0 . 25 mm
the model
from
., Therefore
assume
observations to
thus do not
scale
on
the following
10 :
- vectors(startand
- symbols
- arcs
))
of the
contour
(the Handbook
1978), e.g.
houses. The rules
buildings on three
are not
in
~J~H~~HL,
all
1:2000 .
of
of
the
<
rules"
can
1
and
1 ..
. It can
connected at
the case
estates
None
the
1
to
fences can be
enclosed
a
we have to
lustrates a
that
the
of
the
what s a
for
to
1-----1
1
2
3
Now we
can
I
the
has to have an odd
are some
forms the
of order ..
the
the
The
the
of
marks
for
fence
for
r3=(e2)
to (e1)
r3= (not) (e3
(r4)
e5=number of order
the model
() =value ..
can be
.. If we want
the
However,
these
Table
shows the result from the test above .
Errors of the
order mean
. Errors of the second
as
correct
number
objects
have
been
ects
are not
tot
errors of
1. order
of
as
37
302
305
3
302
47
460
506
46
460
107
1087
1087
1087
The test of the rule for a
All the real
of the
objects
classi
of
result is
been
when
the rules .
The
errors
several
surrounded
I
surrounded
several
I
I
to transform
that allows
at
. It may
has several
a well known
attack the
space
for PC are
be
handled
on
real
P .. ,
I
Center,
1984 ..
(19
for
Carto London
Forbes, R I (19
GIS: Toward the
meeting ASP,
Goldkuhl, G .. ,
1978 ..
S
I
(1984),
.. nr .. 744
1986 ..
Keates I J" S .. ,
Essex, UK, 1982 ..
Weber, Wo, (1987
on
I
)
feature