МЕТОД РАСПОЗНАВАНИЯ ФИГУР С ИСПОЛЬЗОВАНИЕМ

64
-
. . , 634034, , 004.931
! " # . $" " !
$ ! # ! #-
% %% . & " !$
! #-
" " %" % ! $" #.
: #-
", ! #, %" , "% .
This paper provides a brief review on contour-based shape signatures. A new algorithm for shape
recognition based on Fourier descriptors and multilayer neural network is proposed. The paper also presents an analysis on the capabilities of the Fourier-descriptors as the input data for neural networks in
recognition of the complex shapes.
Key words: Fourier descriptors, shape recognition, neural networks, multilayer perceptron
!"#. ! ! % !% ' !.
( ! ! )
!$ !* # – * ! !. +##
" !" # ! #-
[1]. / % #
" *
%% * ' ! !
#.
$% &%! '$!%"#( )#*+$ " " ,"%+$. &2
*2 " # $ #
*2 !: 5
", " , ", " , !; #-*2 (IP) #-*2 " (NIP). / !
" "* * # " !*
" # #" (one-dimensional function), * (olygonal approximation), !! " ! (spatial
interrelation feature), " (moments), " '" (scale-space methods), "
! #" (shape transform domains) [2].
$" # !*
" ", # , "% , ! #-
-
65
. . ". / 5
( #-
) " #%
#".
. , !$ #
f(x,y) Pn ( xn , yn ), n [1, N ] – $
(
) #". / 5
zn=xn+iyn !"
% %, * $ !
#" " " #-!. 5
!
# Pn=zn, n=[1,N].
"% !, : "% !
" 2*, '
* 2*. @
" " " zn " " 2*, "*
$
(
): zn=(xn–xg)+i(yn–yg), g=(xg,yg) – $
#".
. B Rn Pn=(xn,yn), n=[1,N] "
$% C(x0,y0) $% (xn,yn). / " "
#" [2]:
( xn x0 )2 ( yn y0 )2 .
Rn
B $ 2
, " ".
. $"% % %, 5
$ %, % $% [2]:
Tn
§ y yn w ·
arctg ¨ n
¸.
© x n xn w ¹
C w – ' !.
!, "% 2
" : ' "
. " !$
"
, # Mn T n T 0 , 0 – "% %% "%
. 5
% # #
$ . / 2% # !
% # #-!.
-
. B-
" *
!
#-! !" "' " # #" [3, 4]. B-
!"*
" 5##
" #-!$. $, )
!
"% % #% c(t), 5
T
ak
2
c(t ) cos( kZt )dt , bk
T ³0
T
2
c(t ) sin( kZt ) dt , ck
T ³0
ak 2 bk 2
(ak – ; bk – ; ck – #-
).
B-
" %" 2*, '
* 2* )
[2, 3], !
!
#".
*$#%& # * $#/0#(. ! " *
5
: . D2 ! . 1. E! " $
66
!$
! #
5##
"
#-!
% % #
#!
!
" (! #")
. 1. D2 . 2. E! " %% 20 "
"" !$% )
#. D)
" *2
": $
, , , " (. 2).
" ' " !
(
(Moore’s neighbors) [5]. !
" " )
2* (
. 3.
/""% Pn = (xn, yn), n = [1, N], N – " . " % . / $% % !
% ( #). #
! [0, 2!). !, " (!% ! 2! 0) $
$
% #% #-!. 5
% "
!
#. D # : " % ! !
" . " $ " #-!, # $ "
[3]:
§ L ·
M* (t ) M ¨ t ¸ t
© 2S ¹
(J – #; J* – #; L – ).
. 3. /" ! (:
– !$; – )
"" 67
. . y
y
y
x
*
Re(*)
t
t
Im(*)
"
F
t
t
. 4. !
" #-! $
:
– "% ; – #; – #;
– #; – #-!;
– #-!; " – #-
/ !
#-! % % %
# (. 4,–"):
ak
1
S
2S
³M
0
*
(t ) cos(kt ) dt , bk
1
S
2S
³ M (t ) sin(kt ) dt ,
*
ck
ak2 bk2 .
0
" ! #-
" " 2*, '
* 2* "
!" " " %% . 5##
#-! %% !
"$
" #". +
!"
, ! $" # 15–20 5##
. / % !*
20 5##
(
).
! # % % " ', % ! . 5. / #
!
" #. "' *
*2 #: (Nguyen – Widrow)-,
, [6]. +
!"
, 40 – 60 % "
*
'% !
! ' – .
/ % !*
50 "
" %.
68
/% % –
20 %
&"
"% % –
50 %
/"% % –
4 %
. 5. & %% 1+2!"# $/+3%%. , ! !" C# 2008, !$
# !" " ! %% (%"% ), $ "% #% .
5
10 000 5 % '%, % 0,001. 50 ! ! % !" ",
2% ! 18 !$% (. 6). @
' 0,1 %.
5##
" !"
!" " (. 7). " !
" !, !
!
$" #", 2 ! "
5
(, 5, . .) "% . 6. " !$
*: !
", !" . 7,
30 $" ' (
' – 0,15 %).
. 7. ! $" #:
– ; – 69
. . !. !, ! ! # !
! #-
%% . !, #-
%" % 5##
" ' ! ! )
. !
!
$" #" "% *.
& "
1. FOLKERS A., SAMET H. Content-based image retrieval using Fourier descriptors on a logo database //
Proc. of the 16th Intern. conf. on pattern recognition, Quebec (Canada), 11–15 Aug. 2002. Washington:
IEEE Computer Soc., 2002. V. 3. P. 521ದ 524.
2. ZHANG D., LU G. Review of shape representation and description techniques // Pattern Recognition.
2004. V. 37. P. 1–19.
3. NIXON M. Feature extraction and image processing / M. Nixon, A. Aguado. Oxford: Elsevier, 2008.
406 p.
4. PATTERN recognition techniques, technology and applications / Ed. by Peng-Yeng Yin. Croatia: InTech, 2008. 626 p.
5. GHUNEIM A. G. Moore-neighbor tracing // Contour Tracing. 2010. http://www.imageprocessingplace.
com/downloads_V3/root_downloads/tutorials/contour_tracing_Abeer_George_Ghuneim/moore.html.
6. FAUSETT L. V. Fundamentals of neural networks-architectures, algorithms, and applications. Upper
Saddle River: Prentice Hall, 1993. 461 p.
– . $
%
; .: (382-2)70-16-09; e-mail: thangngt.cntt@gmail.com
– 02.11.11