HANDWRITING RECOGNITION IN INTELLIGENT DESIGN...
Transcript of HANDWRITING RECOGNITION IN INTELLIGENT DESIGN...
A R C H I V E S O F M E C H A N I C A L T E C H N O L O G Y A N D A U T O M A T I O N
Vol. 34 no. 3 2014
WOJCIECH KACALAK*, MACIEJ MAJEWSKI**
HANDWRITING RECOGNITION
IN INTELLIGENT DESIGN SYSTEMS
This article presents selected research on the development of complex fundamentals of build-
ing intelligent interactive systems for design of machine elements and assemblies on the basis of
its features described in a natural language. We propose a new method for handwriting recognition
that utilizes geometric features of letters. The article deals with recognition of isolated handwritten
characters using neural networks. As a result of the geometrical analysis, graphical representations
of recognized characters are obtained in the form of pattern descriptions of isolated characters.
Selected parameters of the characters are inputs to the neural network for writing recognition
which is font independent. In this article, we present a new method for off-line natural writing
recognition and also describe our research and conclusions on the experiments.
Key words: human-machine interaction, intelligent automated design, intelligent system,
handwriting recognition, artificial neural networks, computer aided design
1. INTRODUCTION
The presented research [6] involves the development of complex fundamentals
of building intelligent interactive systems for design of machine elements and
assemblies on the basis of its features described in a natural language.
In machine automation technology, the problem of writing recognition is quite
complex [3], and even now there is no single approach that solves it both efficient-
ly and completely in all contexts. In written language recognition processes, an
image containing text must be appropriately supplied and preprocessed. Then the
text must either undergo segmentation or feature extraction. Small processed piec-
es of the text will be the result, and these must undergo recognition by the system.
Finally, contextual information should be applied to the recognized symbols to
verify the result. Artificial neural networks, applied in handwriting recognition,
allow for high generalization ability and do not require deep background
* Prof. dr hab. inż. ** Dr hab. inż.
Department of Mechanical Engineering, Koszalin University of Techno-
logy.
W. Kacalak, M. Majewski 4
knowledge and formalization to be able to solve the written language recognition
problem. Handwriting recognition can be divided by its input method into two
categories: off-line handwriting recognition and on-line handwriting recognition.
For off-line recognition, the writing is usually captured optically by a scanner. For
on-line recognition, a digitizer samples the handwriting to time-sequenced pixels
as it is being written. Hence, the on-line handwriting signal contains additional
time information which is not present in the off-line signal.
PREPROCESSING SUBSYSTEM
Module of searching of
centre of mass
Module of pattern
description
NEURAL NETWORKSUBSYSTEM
for isolated characters
Module for recognizing
isolated characters using artificial
neural networks
Natural writing
recognised text
Natural written
text
Digitization
Binarization
Noise elimination
Thinning
Normalizing
Segmentation
Module of input value
normalization
Module of fuzzy logic
rules
GEOMETRICAL ANALYSIS SUBSYSTEM
for feature extraction by character
Module of detection and extraction of geometrical
features
Module for recognizing words
using artificial neural networks
ANN trainingpatterns of words
(vocabulary verification)
Module for recognizing phrases
using artificial neural networks
ANN training patterns of phrases
(contextual knowledge from linguistics)
FUZZY LOGIC SUBSYSTEM
Letter string
recognition module
NEURAL NETWORK
SUBSYSTEM
for vocabulary and linguistics
Figure 1. Proposed handwriting recognition system [6]
In the proposed new method [7] of natural writing recognition in Figure 1, the
handwritten text is subject to the following preprocessing: digitization, binariza-
tion, noise elimination, thinning, normalizing and segmentation. The next step is
to find the center of mass of the character image. With the center of mass as
a reference point, radiuses are drawn, creating a set of points describing the con-
tour of the character so that its pattern description is made. In the proposed hybrid
system, the pattern description of each isolated character, after the process of input
value normalization and application of letter description rules using fuzzy logic,
are the input signals for probabilistic neural networks for isolated character recog-
nition. The recognized characters are grouped into more quantitative units with the
letter string recognition module, which are coded as binary images of vectors and
then become inputs of the module for recognizing words. The module uses
a 3-layer Hamming neural network [11, 12]. The network of this module uses
a training file containing patterns of words. The recognized vocabulary words
represented by the output neurons are processed by the module for recognizing
phrases which uses the Hamming Maxnet network equipped with a training file
containing phrases built with contextual knowledge from linguistics.
Handwriting recognition in intelligent design systems 5
2. THE STATE OF THE ART
The state of the art of automatic recognition of handwriting at the beginning of
the new millennium is that as a field it is no longer an esoteric topic on the fringes
of information technology, but a mature discipline that has found many commer-
cial uses. On-line systems for handwriting recognition are available in hand-held
computers such as personal digital assistants. Their performance is acceptable for
processing handprinted symbols, and when combined with keyboard entry, a pow-
erful method for data entry has been created.
Off-line systems are less accurate than on-line systems. However, they are now
good enough that they have a significant economic impact on specialized domains
such as interpreting handwritten postal addresses on envelopes and reading courte-
sy amounts on bank checks [1, 2, 3, 9, 10, 13, 19].
The success of on-line systems makes it attractive to consider developing off-
-line systems that first estimate the trajectory of the writing from off-line data and
then use on-line recognition algorithms [15]. However, the difficulty of recreating
the temporal data has led to few such feature extraction systems so far [1].
Research on automated written language recognition dates back several dec-
ades. Today, cleanly machine-printed text documents with simple layouts can be
recognized reliably by OCR software. There is also some success with handwrit-
ing recognition, particularly for isolated handprinted characters and words. For
example, in the on-line case, the recently introduced personal digital assistants
have practical value. Similarly, some online signature verification systems have
been marketed over the last few years and instructional tools to help children learn
to write are beginning to emerge. Most of the off-line successes have come in
constrained domains, such as postal addresses, bank checks, and census forms.
The analysis of documents with complex layouts, recognition of degraded printed
text, and the recognition of running handwriting continue to remain largely in the
research arena. Some of the major research challenges in on-line or off-line pro-
cessing of handwriting are in word and line separation, segmentation of words into
characters, recognition of words when lexicons are large, and the use of language
models in aiding preprocessing and recognition. In most applications, machine
performance is far from being acceptable, although potential users often forget that
human subjects generally make reading mistakes [2, 3, 5].
The design of human-computer interfaces [6, 7, 8, 12] based on handwriting is
part of a tremendous research effort together with speech recognition, language
processing and translation to facilitate communication of people with computers.
From this perspective, any successes or failures in these fields will have an im-
portant impact on the evolution of languages [4, 14].
W. Kacalak, M. Majewski 6
3. DESCRIPTION OF THE METHOD
The proposed system [6] attempts to combine two methods for natural writing
recognition, neural networks and preprocessing for geometric features extraction.
The system consists of the preprocessing subsystem, geometrical analysis subsys-
tem, fuzzy logic subsystem, neural network subsystem for isolated characters as
well as neural network subsystem for vocabulary and linguistics, as shown in Fig-
ure 2. The motivation behind that preprocessor is to reduce the dimensionality of
the neural network input. However, another benefit given by the preprocessor is
immunity against image translation, because all the information is relative to the
image's center of mass.
The developed geometrical analysis is based on the processing of the images of
letter shapes into their graphical representations in the form of pattern descriptions.
The process of the geometrical analysis begins with determining of the center of
mass of a letter with a gravity method in order to find the initial point of the analy-
sis. The next step of the algorithm is based on drawing radiuses from the initial
point, the lengths of which are equal to the length of the line segment created by
the initial point and the point on the letter furthest from this point. The creation of
a circle of that radius makes it visible that the analysis covers the whole letter. The
precision of this geometrical analysis method is proportional to the number of
radiuses.
Where the radiuses intersect with the letter, points are obtained, which makes it
possible to obtain the measures of the line segment created by the initial point and
the letter intersection point. The lengths of the created line segments obtained are
represented in the form of pattern descriptions of isolated characters which are
inputs of the probabilistic neural network. Geometrical analyses of characters for
exemplary letters are shown in Figure 3, which are also used by probabilistic neu-
ral networks to recognize isolated characters.
The architecture of modified probabilistic neural networks for recognition of
pattern descriptions of isolated characters is shown in Figure 4. It is composed of
interconnected neurons organized in successive layers. The probabilistic neural
network was first introduced by Specht [16, 17, 18]. The probabilistic network
consists of input, pattern, summation and output layer.
In the proposed handwriting recognition system, because of the binary input
signals, the Hamming neural network is chosen for both the word recognition and
phrase recognition [11, 12] as shown in Figure 2. The network directly realizes the
one-nearest-neighbor classification rule.
Handwriting recognition in intelligent design systems 7
PR
EP
RO
CE
SS
ING
SU
BS
YS
TE
M
natural written text
Digitization, Binarization,Noise elimination, ThinningNormalizing, Segmentation
M for ing ofe of m
searchodulecent r ass of
an isolated character
M for of g seometrical featureodule detection and extraction
Phrase recognition moduleusing neural networks
Word recognition moduleusing neural networks
NE
UR
AL N
ET
WO
RK
SU
BS
YS
TE
M fo
r vo
cab
ula
ry a
nd
lin
gu
istics
GEOMETRICAL ANALYSIS
SUBSYSTEM
Network input:N=a*b
Classification to phrase classes linguistics) (Contextual knowledge from
Network input:N=26*a
OUTPUT:binary images ofthe recognised
words
Cla
ssific
ation
to
word
cla
sse
s(V
ocab
ula
ry p
atte
rns)
NETWORK INPUT:binary images
as isolatedcomponents of
the textcreated by
the letter stringrecognition
module
binary image ofrecognised
word
NEURAL NETWORK SUBSYSTEM for isolated characters
Natural writing recognised text
Isolated character recognition moduleusing artificial neural networks
Letter string recognition module
Ob
raz b
ina
rny fra
zy
Ob
raz b
inarn
y r
ozpozna
ne
j fr
azy
WYRAZ 1WYRAZ 2WYRAZ 3
WYRAZ b
Numer wyrazu
b
a
WYRAZ 1WYRAZ 2WYRAZ 3
WYRAZ b
b
RO
ZP
OZ
NA
WA
NIE
r1 r2
r3
r4
r5r6r7
r8
r9
r10
r11
r12
f=4
Lo
Lfi
r
xc
yc
r1r2
r3
r4
r5
r6r7
r8
r9
r10
r11
r12
f=4
Lo
Lfi
r
xc
yc
r1
r2
r3
r4
r5
r6r7
r8
r9
r10
r11
rn
f=1
f=7
Lo
Lfi
r
xc
yc
r1 r2
r3
r4
r5r6r7
r8
r9
r10
r11
r12
f=4
Lo
Lfi
r
xc
yc
Δro2
s
o1p1
p3
p7
pn
s(x , y )c c
r1
r2
r3
r4
r5
r6
r7
r8
r9
r10
r11
rn
f=1f=12
f=7
Lo Lfi r
xc
yc
o2
s
o1
s(x , y )c c
p2
p8
p11
0
102030
405060
708090
100
1 2 3 4 5 6 7 8 9 10 11 120
102030405060708090
100
1 2 3 4 5 6 7 8 9 10 11 12
H(?r ), H(? ), H(? )fi i bi
Geometrical analysis results:
Input layer
Statepatternlayer
Summation layer
Pi
P1
(Cluster decision)
Stateclass
1
Stateclass
2
Stateclass
i
Stateclass
K
1
1
i
m
f
f
f
f
f
f
class ki
i=1, ..., K
f
(x)
(x)
(x)
(x)
(x)
(x)
2
1f
i
1f
f
f
K
Nf{ 1 2 K X , X , ..., X }
Parameter input:
Weights:
x1
x2
xj
xp
0
1
0
1
( )å=
-=
=-=
=
n
1j
2ijj
i
i
wx
wx
w,xd
å=
=n
1i
ijij xwy
å=
=n
1i
ii22 xwy
å=
=n
1i
ii11 xwyK,,2,1i KÎ
å=
=n
1i
ipip xwy
( )
( ) ( )
( ) ( )[ ]twtx
tηtw
1tw
j
j
j
-´
´+=
=+
å=
=n
1i
i3i3 xwy
å=
=K
1i
iNN
Pi (x
)
wp3
wpn
wp1
P i(x)
P i(x)
Pi (x)
PK (x)
Pi (x)
PK
P2
Figure 2. Methodology of the proposed system for handwriting recognition [6]
W. Kacalak, M. Majewski 8
f=1f=2
f=3f=4
r1
r2
r3
r4
r5
r6r7
r8
r9
r10
r11
r12
f=5
f=6f=7
f=8
f=9f=10
f=11
f=12
r1
r2
r3
r4
r5
r6
r7
r8
r9
r10
r11
r12
f=1
f=3
r1r12
r1r12
f=12
r10
r1
r2
r3
r4
r5
r6r7
r8
r9
r10
r11
r12
f=2
f=3
f=4fnf1 rr ¸ rf1
rfn
r1
rf
rfx
r1
r2
r3
r4
r5
r6
r7
r8
r9
r10
r11
r12
fnf2f1f llll +++= K
f=1
f=3
fnf1 rr ¸
rf1rfn
r1
r2
r3
r4
r5
r6r7
r8
r9
r10
r11
r12
lf1
lf2lf3
f=2
f=3
f=4fnf1 rr ¸ rf1
rfn
Δr
r1
r2
r3
r4
r5
r6r7
r8
r9
r10
r11
r12
f=2
f=3
f=4fnf1 rr ¸ rf1
rfn
Δr
r1
r2
r3
r4
r5
r6
r7
r8
r9
r10
r11
r12
f1ff rrΔr -= +
f=1f=12
Δr
f=7
Figure 3. Geometrical analyses of characters: A) determination of the center of the mass for exem-
plary letter s; B) determination of intersection points of the letter and the radiuses for exemplary
letter o; C) measurement of the length of line segments l created by letter points in fragments f for
letter e; D) summation of measurements in fragments f containing n radiuses for letter l; E) meas-
urement of the length of line segments of each radius for exemplary letter e and letter l (F); G)
measurements of differences of the radius lengths in each fragment f for exemplary letter e and
letter l (H)
Handwriting recognition in intelligent design systems 9
Input layerState
pattern layer
Summation layer
1
Pi
P1
P2
PK
(Clusterdecision,
competitive)output layer
Letterclass
1
Letterclass
2
Letterclass
i
Letterclass
K
1
1
1
2
1
N
2
2
2
2
N
i
i
m
i
N
f
f
f
f
f
f
class ki
i=1, ..., K
f
X x x xi 1 2 p=[ , , ..., ]Ti i i
1
1(x
)
1
2(x)
1
N (x)1
2
1(x
)
2
2(x)
2
N (x)2
i
1(x
)
i
m(x)
iN (x)i
K
1(x
)
K
2(x)
K
N (x)KP
i (x)x1
2
1f
x2
i
1f
xj
K
1f
K
2f
K
K
Nf
xp
{ 1 2 K X , X , ..., X }
wj j = xi i
Parameter input:
Weights:
x1
0
1
Normalization functionfor each parameter value
Input signal vectorsanalyzed data of
describing geometrical features
of characters
Neural network inputs: vectors with encoded character feature values
( )å=
-=
=-=
=
n
1j
2ijj
i
i
wx
wx
w,xd
å=
=n
1i
ijij xwy
å=
=n
1i
ii22 xwy
å=
=n
1i
ii11 xwyK,,2,1i KÎ
wp3
wpn
wp1
å=
=n
1i
ipip xwy
y1
yj
yp
y2
y3
P i(x)
P i(x)
Pi (x)
PK (x)
Pi (x)
( ) ( )
( ) ( )[ ]twtx
tηtw
1tw
j
j
j
-´
´+=
=+
å=
=n
1i
i3i3 xwy
å=
=K
1i
iNN
r1
r2
r3
r4
r5
r6r7
r8
r9
r10
r11
rn
Lo
Lfi
r
xc
yc
s
P1
P3
P7
Pn
s(x , y )c c
Cm
C
Z
rfi rfi rfi
r1
r2
r3
r4
r5
r6
r7
r8
r9
r10
r11
rn
Lo
Lfi
ss(x , y )c c
P2
P8
P11
Cm
C
yc
xc
Z
rm
rfirfi
rfi
0
1
0
1
Figure 4. The architecture of the probabilistic neural network for recognition of pattern descriptions
of isolated characters
4. EXPERIMENTAL RESULTS
The research on the developed method concerns the ability of the neural net-
work to learn to recognize specific letters. The neural networks are trained with
the model of isolated written language characters. Several geometrical analyses of
isolated characters and their pattern descriptions were realized (Figure 5), which
made it possible to draw significant conclusions and apply them in the proposed
algorithms. The ability of the neural network to learn to recognize specific charac-
ters depends on the type of parameters of geometrical features. The specified type
of parameters enables the network to minimize the error so that it can work more
efficiently. Based on the research, the following parameters are the most signifi-
cant for the recognition, as shown in Figure 6.
W. Kacalak, M. Majewski 10
0
20
40
60
80
100
1 2 3 4 5 6 7 8 9 10 11 12r
rad
ius
len
gth
[%]
0
20
40
60
80
100
1 2 3 4 5 6 7 8 9 10 11 12r
rad
ius
len
gth
[%]
0
20
40
60
80
100
1 2 3 4 5 6 7 8 9 10 11 12r
rad
ius
len
gth
[%]
0
20
40
60
80
100
1 2 3 4 5 6 7 8 9 10 11 12r
rad
ius
len
gth
[%]
0
20
40
60
80
100
1 2 3 4 5 6 7 8 9 10 11 12r
rad
ius
len
gth
[%]
0
10
20
30
40
50
60
70
80
1 2 3 4 5 6 7 8 9 10 11 12r
rad
ius
len
gth
[%]
0
20
40
60
80
100
1 2 3 4 5 6 7 8 9 10 11 12r
rad
ius
len
gth
[%]
0
20
40
60
80
100
1 2 3 4 5 6 7 8 9 10 11 12r
rad
ius
len
gth
[%]
0
20
40
60
80
100
1 2 3 4 5 6 7 8 9 10 11 12r
rad
ius
len
gth
[%]
0
20
40
60
80
100
1 2 3 4 5 6 7 8 9 10 11 12r
rad
ius
len
gth
[%]
0
20
40
60
80
100
120
1 2 3 4 5 6 7 8 9 10 11 12r
rad
ius
len
gth
[%]
0
20
40
60
80
100
120
1 2 3 4 5 6 7 8 9 10 11 12r
rad
ius
len
gth
[%]
0
20
40
60
80
100
1 2 3 4 5 6 7 8 9 10 11 12r
rad
ius
len
gth
[%]
0
20
40
60
80
100
120
1 2 3 4 5 6 7 8 9 10 11 12r
rad
ius
len
gth
[%]
0
20
40
60
80
100
1 2 3 4 5 6 7 8 9 10 11 12r
rad
ius
len
gth
[%]
0
20
40
60
80
100
1 2 3 4 5 6 7 8 9 10 11 12r
rad
ius
len
gth
[%]
0
20
40
60
80
100
1 2 3 4 5 6 7 8 9 10 11 12r
rad
ius
len
gth
[%]
0
10
20
30
40
50
60
70
1 2 3 4 5 6 7 8 9 10 11 12r
rad
ius
len
gth
[%]
0
20
40
60
80
100
1 2 3 4 5 6 7 8 9 10 11 12r
rad
ius
len
gth
[%]
0
20
40
60
80
100
120
1 2 3 4 5 6 7 8 9 10 11 12r
rad
ius
len
gth
[%]
0
20
40
60
80
100
1 2 3 4 5 6 7 8 9 10 11 12r
rad
ius
len
gth
[%]
0
20
40
60
80
100
1 2 3 4 5 6 7 8 9 10 11 12r
rad
ius
len
gth
[%]
0
20
40
60
80
100
1 2 3 4 5 6 7 8 9 10 11 12r
rad
ius
len
gth
[%]
0
20
40
60
80
100
1 2 3 4 5 6 7 8 9 10 11 12r
rad
ius
len
gth
[%]
0
20
40
60
80
100
1 2 3 4 5 6 7 8 9 10 11 12r
rad
ius
len
gth
[%]
Figure 5. Geometrical analysis and pattern description of isolated characters
r1r2
r3
r4
r5
r6r7
r8
r9
r10
r11
r12
f=4
Lo
Lfi
r
xc
yc
r1
r2
r3
r4
r5
r6
r7
r8
r9
r10
r11
rn
f=1f=12
f=7
Lo Lfi r
xc
yc
r1
r2
r3
r4
r5
r6r7
r8
r9
r10
r11
rn
f=1
f=7
Lo
Lfi
r
xc
yc
r1 r2
r3
r4
r5r6r7
r8
r9
r10
r11
r12
f=4
Lo
Lfi
r
xc
yc
r1 r2
r3
r4
r5r6r7
r8
r9
r10
r11
r12
f=4
Lo
Lfi
r
xc
yc
Δro2
s
o1p1
p3
p7
pn
s(x , y )c c
o2
s
o1
s(x , y )c c
p2
p8
p11
Figure 6. Geometrical analyses of characters for recognition of handwriting
using neural networks
Handwriting recognition in intelligent design systems 11
5. CONCLUSIONS AND PERSPECTIVES
The presented research involves the development of complex fundamentals of
building intelligent interactive systems for design of machine elements and assem-
blies on the basis of its features described in a natural language. The advantages of
this new method of natural writing recognition are flexibility with regards to writ-
ing style, geometrical analysis enabling font independent character recognition,
possibility of application of other types of neural networks, extension of the range
of geometrical analysis and other possibilities for further development.
ACKNOWLEDGEMENT
This project was financed from the funds of the National Science Centre (Po-
land) allocated on the basis of the decision number DEC-2012/05/B/ST8/02802.
REFERENCES
[1] Artieres T., Gauthier N., Gallinari P., Dorizzi B., A Hidden Markov Models combination
framework for handwriting recognition, International Journal on Document Analysis and
Recognition, 2003, vol. 5, no. 4, p. 233–243.
[2] Besner D., Humphreys G.W., Basic Processes in Reading: Visual Word Recognition, New
Jersey, Lawrence Erlbaum Associates 1991.
[3] Bishop C.M., Neural Networks for Pattern Recognition, New York, Oxford Univ. Press
2004.
[4] Bradford R., Bradford R.B., An Introduction to Handwriting Examination and Identifica-
tion, Chicago, Nelson-Hall Publishers 1992.
[5] Dori D., Bruckstein A., Shape, Structure and Pattern Recognition, New Jersey, World Sci-
entific Publishing Co. 1995.
[6] Kacalak W., Majewski M., Effective Handwriting Recognition System using Geometrical
Character Analysis Algorithms, in: 19th International Conference on Neural Information
Processing ICONIP2012, Doha, Qatar, 12–15 November 2012. Lecture Notes in Computer
Science 7666, Part 4, Springer 2012, p. 248–255.
[7] Kacalak W., Majewski M., New Intelligent Interactive Automated Systems for Design of
Machine Elements and Assemblies, in: 19th International Conference on Neural Information
Processing ICONIP2012, Doha, Qatar, 12–15 November 2012. Lecture Notes in Computer
Science 7666, Part 4, Springer 2012, p. 115–122.
[8] Kacalak W., Majewski M., Zurada J.M., Intelligent e-learning systems for evaluation of
user's knowledge and skills with efficient information processing. Lecture Notes in Artificial
Intelligence 6114, Springer 2010, p. 508–515.
[9] Kavallieratou E., Fakotakis N., Kokkinakis G., An unconstrained handwriting recognition
system, International Journal on Document Analysis and Recognition, 2002, vol. 4, no. 4,
p. 226–242.
[10] Li Z.C., Suen C.Y., The partition-combination method for recognition of handwritten char-
acters, Pattern Recognition Letters, 2000, vol. 21, no. 8, p. 701–720.
W. Kacalak, M. Majewski 12
[11] Lippman R., An Introduction to Computing with Neural Nets, IEEE Transactions on Acoustic,
Speech, and Signal Processing, IEEE Signal Processing Society, 1987, vol. 4, no. 3, p. 4–22.
[12] Majewski M., Zurada J.M., Sentence recognition using artificial neural networks,
Knowledge-Based Systems, 2008, vol. 21, issue 7, p. 629–635.
[13] Mitiche A., Lebidoff M., Pattern Classification by a Condensed Neural Network, Neural
Networks, 2001, vol. 14, no. 4–5, p. 575–580.
[14] Mori S., Nishida H., Yamada H., Optical Character Recognition, New York, Wiley 1999.
[15] Nishida H., An Approach to Integration of Off-Line and On-Line Recognition of Handwrit-
ing, Pattern Recognition Letters, 1995, vol. 16, no. 11, p. 1213–1219.
[16] Parzen E., On estimation of a probability density function and mode, The Annals of Mathe-
matical Statistics, 1962, vol. 33, no. 3, p. 1065–1076.
[17] Specht D.F., A general regression neural network, IEEE Transactions on Neural Networks,
1991, vol. 2, no. 6, p. 568–576.
[18] Specht D.F., Probabilistic neural networks, Neural Networks, 1990, vol. 3, no. 1, p. 109–118.
[19] Zhou J., Krzyzak A., Suen C.Y., Verification-a Method of Enhancing the Recognizers of
Isolated and Touching Handwritten Numerals, Pattern Recognition, 2002, vol. 35, no. 5,
p. 1179–1189.
ROZPOZNAWANIE PISMA ODRĘCZNEGO
W INTELIGENTNYCH SYSTEMACH PROJEKTOWANIA
S t r e s z c z e n i e
W artykule przedstawiono wybrane prace badawcze dotyczące podstaw budowy inteligent-
nych systemów interakcji do projektowania elementów i zespołów maszyn na podstawie ich cech
opisywanych w języku naturalnym. Zaproponowano nową metodę rozpoznawania pisma odręcz-
nego, w której wykorzystano geometryczne cechy znaków. Artykuł dotyczy rozpoznawania izo-
lowanych znaków pisma odręcznego za pomocą sieci neuronowych. W wyniku analizy geome-
trycznej otrzymuje się reprezentacje graficzne rozpoznawanych znaków w postaci opisów wzor-
ców pojedynczych znaków. Wybrane parametry znaków stanowią wejścia sieci neuronowej do
rozpoznawania pisma niezależnego od kroju. W artykule przedstawiono nową metodę rozpozna-
wania pisma naturalnego, a także opisano badania i podano wnioski wynikające z eksperymentów.
Słowa kluczowe: interakcja człowiek–maszyna, inteligentne zautomatyzowane projektowanie,
inteligentny system, rozpoznawanie pisma odręcznego, sieci neuronowe,
CAD