Outline Representation KFUPM KFUPM of Fonts Using Genetic - - PowerPoint PPT Presentation
Outline Representation KFUPM KFUPM of Fonts Using Genetic Approach M. Sarfraz sarfraz@kfupm.edu.sa Information & Computer Science Department Information & Computer Science Department King Fahd Fahd University of Petroleum &
KFUPM KFUPM Slide Slide 1 1 WSC7 WSC7
Using Genetic Approach
sarfraz@kfupm.edu.sa Information & Computer Science Department Information & Computer Science Department King King Fahd Fahd University of Petroleum & Minerals University of Petroleum & Minerals Dhahran, Saudi Arabia Dhahran, Saudi Arabia
KFUPM KFUPM Slide Slide 2 2 WSC7 WSC7
Problem
Solution
Conventional Method
Automating Procedure: Various Steps
Various Steps
Results
Conclusion & Research Issues Research Issues
KFUPM KFUPM Slide Slide 3 3 WSC7 WSC7
Outline of Outline of Fonts Fonts
KFUPM KFUPM Slide Slide 4 4 WSC7 WSC7
Bitmap
Example: Fonts
Outline
KFUPM KFUPM Slide Slide 5 5 WSC7 WSC7
1.
Scan the original character drawn on paper.
2.
Identify Significant Points.
3.
Fit Bezier to Significant Points.
4.
Match with original shape, if undesirable goto step 2.
KFUPM KFUPM Slide Slide 6 6 WSC7 WSC7
Drawbacks
Identifying Significant Points:
Matching with original character is
difficult.
Human intervention is inevitable. Variation from case to case
KFUPM KFUPM Slide Slide 7 7 WSC7 WSC7
1.
Scan the original character.
2.
Identify Significant Points.
3.
Filtering the Contour
4.
Detection of More Significant Points using GAs and NUCS.
5.
Match with original shape, if undesirable goto goto step 4.
KFUPM KFUPM Slide Slide 8 8 WSC7 WSC7
KFUPM KFUPM Slide Slide 9 9 WSC7 WSC7
Piece 2 Piece 1
Example 1
KFUPM KFUPM Slide Slide 10 10 WSC7 WSC7
Piece 2 Piece 1 Piece 3 Piece 4 Piece 5
Example 2
KFUPM KFUPM Slide Slide 11 11 WSC7 WSC7
Each piece consists of points pi=(xi,yi);
i=1,…,N
963+87+87+8 7+87 = 1311 5
Thamar
1522+116 = 1638 2
Lillah
# of Boundary Points # of Pieces
Figure
KFUPM KFUPM Slide Slide 12 12 WSC7 WSC7
Corners: High curvature points in planar
curves.
If α is measure of curvature then corners are
characteristics contour points whose α value is less than some predefined threshold value i.e. αmax.
Input: Sequence of points pi =(xi, yi),
i=1,2…,N, on contour.
Candidate Points: testing back and forth of
a point pi, for potential candidacy.
KFUPM KFUPM Slide Slide 13 13 WSC7 WSC7
a b c
A
p p- p+
KFUPM KFUPM Slide Slide 14 14 WSC7 WSC7
Pass-1: Each candidate point pi may have
multiple associated cornality values…. minimum value is taken as the alpha value
Pass-2: Selecting most suitable Point
among the adjacent candidates
Why not Pass – –1 algorithm? 1 algorithm?
KFUPM KFUPM Slide Slide 15 15 WSC7 WSC7
KFUPM KFUPM Slide Slide 16 16 WSC7 WSC7
Conics
Bezier Cubic Cubic
UBS
NUBS
NURBS
NUCS
Others
KFUPM KFUPM Slide Slide 17 17 WSC7 WSC7
NUCS represent Non-uniform Cubic Spline NUCS has ideal geometric properties NUCS is computationally efficient A NUBS is represented as: NUCS represent Non-uniform Cubic Spline NUCS has ideal geometric properties NUCS is computationally efficient A NUBS is represented as:
( ) ( )
( ) ( ) ( )
, F W V F t | P
i i i i i i i i i i t , t
i i
1 3 2 2 3
1 3 1 3 1
1
+
+ − + − + − =
+
θ θ θ θ θ θ ( )
( ) ( )
( )
, h / t t t | t
i i t , t i i
i i− = = ≡
+1θ θ θ
where and
, D h F V
i i i i
3 1 + = . D h F W
i i 1 i i 1
3 1
+ + −
=
The NUCS The NUCS The NUCS
KFUPM KFUPM Slide Slide 18 18 WSC7 WSC7
0 1 0 1 -------------------- 0 1 a b Knots
KFUPM KFUPM Slide Slide 19 19 WSC7 WSC7
Crossover Rate
0.7
Mutation Rate
0.001
120
Population size
30
Knot Ratio
0.3
KFUPM KFUPM Slide Slide 20 20 WSC7 WSC7
Population for weights Generations for weights Decimation Gene Length = No. of Points after decimation
KFUPM KFUPM Slide Slide 21 21 WSC7 WSC7 Algorithm DESIGN Algorithm DESIGN Algorithm DESIGN
Input Control Parameters START Extract the Contour Get Digitized Image Detect Significant Knots Fit NUCS Curve using GA (iterative process) STOP
KFUPM KFUPM Slide Slide 22 22 WSC7 WSC7
Curve Fitting
Input Data START Initialize Population Calculate Fitness Calculate NUCS Fitting Do Selection Do Crossover & Mutation No Yes Max Generations Exhausted?
KFUPM KFUPM Slide Slide 23 23 WSC7 WSC7
Pseudo code for Font Design System Pseudo code for Font Design System
Algorithm (Font_design)
Initialize Population While i< = Max_Gen Do For j <= Pop_size Initialize weights For k <= weights_gen For m<=pop_size Calculate NUCS Calculate AIC Selection Crossover Mutation End For Store Best Solution End For Make new Population End For Selection Crossover Mutation i = i + 1 Conserve Corner Points End Do
KFUPM KFUPM Slide Slide 24 24 WSC7 WSC7
Have Total Generations Exhausted STOP for j<=population size Initialize Weights Population for m<=weights population Calculate NUCS Calculate AIC Do Selection Do Crossover Do Mutation Store Best Result Make New Population Do Selection Do Crossover Do Mutation for k<=weights generations Make New Population i = i + 1
No
k>weights generations
No Yes Yes
while i<=MaxGenerations Initialize Population Conserve CornerPoints in the Population Start
KFUPM KFUPM Slide Slide 25 25 WSC7 WSC7
Bitmap Image Bitmap Image After Boundary Detection After Boundary Detection
KFUPM KFUPM Slide Slide 26 26 WSC7 WSC7
After Applying Corner Detection After Applying Corner Detection
KFUPM KFUPM Slide Slide 27 27 WSC7 WSC7
After NUCS Fitting After NUCS Fitting
KFUPM KFUPM Slide Slide 28 28 WSC7 WSC7
Bitmap Image Bitmap Image After Boundary Detection After Boundary Detection
KFUPM KFUPM Slide Slide 29 29 WSC7 WSC7
After Applying Corner Detection After Applying Corner Detection and NUCS Fitting and NUCS Fitting
KFUPM KFUPM Slide Slide 30 30 WSC7 WSC7
A Font Design Method
Equally Applicable for Hand-
drawn Shapes Shapes
Efficient and Accurate Solution?
Better than Conventional Method
Full Automation?
Visually Pleasant Results
Research Issues
Corner Detection
Segmentation
Any Other Curve Model?
KFUPM KFUPM Slide Slide 31 31 WSC7 WSC7
KFUPM KFUPM Slide Slide 32 32 WSC7 WSC7