КОДИРОВАНИЕ КОНТУРОВ, ПРЕДСТАВЛЕННЫХ КРИВЫМИ БЕЗЬЕ, В ЗАДАЧАХ НЕЙРОСЕТЕВОЙ КЛАССИФИКАЦИИ

  • Published on
    29-Mar-2017

  • View
    213

  • Download
    1

Embed Size (px)

Transcript

  • ISSN 1028-9763. , 2004, 3

    17

    621.8:681.5

    ..

    , , Abstract: The article is devoted to describe method and algorithms of coding the contours represented with Bezier curves for neural network pattern recognition. Because a contour description can consist arbitrary number and sequence of the curves, neural networks utilization requires extra coding. For the coding there are used histograms based on parametric notion of curves in the contour. There are proposed limitations on Bezier curves that allow significantly reduce the length of obtained codes. Efficiency of developed method is shown with example of classification of plane figures. Key words: pattern recognition, coding of the contours, neural networks. : , , . , . . , . . : , , . : , , . , . . , . . : , , . 1.

    ,

    , .

    .

    , , ,

    , .

    ,

    . (,

    ) .

    .

    ,

    .

    . ,

    , , , .

    ,

    , , .

  • ISSN 1028-9763. , 2004, 3

    18

    , .

    .

    2.

    ,

    [1]. ,

    , . , ,

    ,

    . , ,

    .

    (anchor) ,

    , , , (control) ,

    . .

    N- N+1 .

    ,

    , , .

    , , TrueType .

    :

    3223 tDtt)-(1C3t)-(1tB3t)-(1AP(t) +++= ,

    A ; D , B C .

    AB A; CD D. t 0

    1. 0=t ( ) AtP = , 1=t ( ) DtP = . .

    5,0=t , . , :

    1) ;

    2) ;

    3) ;

    4) ;

    5) , , ,

    ;

    6) , ,

    .

    ,

    ,

    .

  • ISSN 1028-9763. , 2004, 3

    19

    .

    [2].

    .

    ( ).

    :

    1) ;

    2) ( )

    .

    ,

    .

    3.

    .

    . ,

    , ,

    , .

    .

    (

    , ).

    .

    ,

    ,

    . ,

    ,

    ,

    ( ).

    , , ,

    :

    1)

    ;

    2) , , ,

    , .

    , 20 .

    (

    ) , , . ,

    0,05

  • ISSN 1028-9763. , 2004, 3

    20

    . ,

    .

    .

    .

    . , , ,

    ,

    , .

    , ,

    ,

    .

    ,

    , . , .

    , ,

    , ,

    .

    3.1.

    0P 2P ,

    1P . 1020 , PPcPPb ==rr

    .

    , bcdrr= ,

    . ,

    N N2 .

    . 1.

    ) )

    ) )

    b

    c

    P1

  • ISSN 1028-9763. , 2004, 3

    21

    ,

    ,

    . 1, 1), 1) 1), 1)

    . . 1) br

    cr

    ,

    .

    1+= iii bbgrr

    ),( 1+= iii bbrr

    , i

    . k4

    ( kigd iiii ,1,,, = ), k . ,

    .

    , ,

    .

    k4 , ,

    .

    .

    ,

    , , .

    . 1) 1)

    (, ), ,

    . 1) 1) ,

    d . , ().

    , gd ,,

    .

  • ISSN 1028-9763. , 2004, 3

    22

    24 mn . .

    ,

    , ,

    , .

    1: , ,

    :

    ,

    , , ,

    , ( );

    ,

    , ,

    , ,

    .

    ,

    , .

    ,

    . ,

    .

    1 . 2, ,

    ( ).

    . 2.

    )

    )

    . 2) ,

    , . 2) , .

  • ISSN 1028-9763. , 2004, 3

    23

    2) , , ,

    .

    . , ,

    n m 10,

    106, .

    , ,

    .

    , d . ,

    . ,

    . , g

    , d . , nm > .

    ,

    , . ,

    22

  • ISSN 1028-9763. , 2004, 3

    24

    : , )sin(dA = ,

    .

    m9 , . . 3

    ,

    g . . 3)

    0P .

    . 3. ,

    )

    )

    .

    NbbbB K = },...,{ 1 :

    1) maxR

    0P ;

    2) K

    KR /max= ;

    3)

    ,0

    11

    ;1