A neural network model for complex handwriting generation movements

apricotpigletAI and Robotics

Oct 19, 2013 (3 years and 5 months ago)

50 views

Mahmoud
Ltaief

IGS 2011
-

L
IVE

A
QUA

C
ANCUN
, MEXICO

1

Re
search

G
roup

on

I
ntelligent

M
achines


A neural network model for
complex

handwriting

generation

movements


Mahmoud
Ltaief

Hala
Bezine

Adel

M.
Alimi

Mahmoud.ltaief
@
ieee.org

h
a
la.bezine@ieee.org

adel.alimi@ieee.org

National School of

Engineers of Sfax

REsearch Group on

Intelligent Machines

Extraction of the curvilinear velocity signal V(t)

Generation of the Beta profiles with the kinematic parameters (t0,
tc
,
t1, p, q, H) for each stroke

Extraction of the associated static parameters (a, b, X
0
, Y
0
, θ
0
) related to
the elliptic shape

On
-
line handwriting script

Production of handwriting requires a hierarchically organized flow of
information undergoing a series of transformations.

Mahmoud
Ltaief

IGS
2011
-

L
IVE

A
QUA

C
ANCUN
,
MEXICO 2

Neural network for handwriting generation

X(t)

Y(t)

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

Timing network

Input layer

Hidden layers

Output layer

Parameters of Beta elliptic model

The

network

was

trained

using

the

back

propagation

algorithm

to

produce

the

coordinates

of

the

initial

script
.

A

timing

network

is

also

used

to

prepare

the

input

data
.

1.7 10
-
5

4.3 10
-
4

5.6 10
-
3

2.2 10
-
2

10

1.2 10
-
3

5.5 10
-
2

4.4 10
-
1

5.3

5

4

3

2

1


Nh

Nn

Mean error table

The

structure

(number

of

hidden

layers

Nh

and

number

of

neurons

per

hidden

layer

Nn)

of

the

proposed

neural

network

is

defined

after

several

simulation

experiments

studying

the

effect

of

various

network

parameters

structure

(Nh

and

Nn)

on

the

training

performance
.

Original letter

Generated letter

Simulation results