Mr. Sovann EN
5th
Year Engineering student
Dept. Computer Science & Communication
Institute of Technology of Cambodia
Phnom Penh, Cambodia
Khmer OCR Using Generic Fourier
Descriptor
Content
Khmer OCR Using Generic Fourier Descriptor
2
 Introduction
 Khmer OCR System
 Pre-processing
 Segmentation
 Feature Extraction
 Recognition Process
 Generic Fourier Descriptor In detail
 Future work
Introduction
 Optical Character Recognition (OCR) is the
electronic conversion of scanned images into machine-
encoded text.
 OCR makes it possible to edit the text, to search for
a word or a phrase, to display or to print a copy free of
scanning artifacts and so on.
3
Khmer OCR Using Generic Fourier Descriptor Back
Introduction
 The ideas started prior to World War II.
Throughout the years, many reliable commercial and
academic prototypes have been developed in many
natural languages.
 Still, due to the lacked effort in Khmer OCR, there
is no reliable Khmer OCR software.
4
Khmer OCR Using Generic Fourier Descriptor Back
Introduction
 The ideas started prior to World War II.
Throughout the years, many reliable commercial and
academic prototypes have been developed in many
natural languages.
 Still, due to the lacked effort in Khmer OCR, there
is no reliable Khmer OCR software.
5
Khmer OCR Using Generic Fourier Descriptor Back
Introduction
 The ideas started prior to World War II.
Throughout the years, many reliable commercial and
academic prototypes have been developed in many
natural languages.
 Still, due to the lacked effort in Khmer OCR, there
is no reliable Khmer OCR software.
6
Khmer OCR Using Generic Fourier Descriptor Back
Introduction
7
Khmer OCR Using Generic Fourier Descriptor Back
 The current system is based on Mr. Vanna Kruy’s
work for his master degree at Waseda University.
 The Objective is to produce a reliable Khmer OCR
system which is independent of Font and Size.
Khmer OCR System
Khmer OCR Using Generic Fourier Descriptor
 OCR System consist of four majors stages :
8
 Pre-processing
 Segmentation
 Feature Extraction
 Post processing
Pre-processing
Khmer OCR Using Generic Fourier Descriptor
9
 Pre-processing aims to produce data that are easy
for the OCR systems to operate accurately.
 The main objectives of pre-processing are :
 Binarization
 Noise removal
Binarization
10
 Image linearization (thresholding) refers to the
conversion of a gray-scale image into a binary image.
Khmer OCR Using Generic Fourier Descriptor
Binarization of Input Image
Salt-Pepper Noise removal
11
Salt-and-pepper noise is a kind of noise which is
usually caused by small unnecessary dots produced by
either the scanner or the source document itself.
Khmer OCR Using Generic Fourier Descriptor Back
Particle removal
Segmentation
12
 Segmentation aims to produce each component to
be recognized by the system.
 The process is to separate the text of a page into
each separate line, then to separate each line into
Vertical Component, and finally produce each
independent symbol.
Khmer OCR Using Generic Fourier Descriptor Back
Connected Component Analysis
13
 Connected Component is a group of pixels
accumulating together to form a shape.
 The process is to separate the text of a page into
each separate line, then to separate each line into
Vertical Component, and finally produce each
independent symbol.
Khmer OCR Using Generic Fourier Descriptor Back
14
Example using CCA
Khmer OCR Using Generic Fourier Descriptor Back
Word Segmentation using CCA
15
Example using CCA
Main part :
Sup script :
Sub script :
Ccdown :
Khmer OCR Using Generic Fourier Descriptor Back
Feature extraction
16
 In feature extraction stage, each character is
represented as a feature vector which becomes its
identity.
The major goal of feature extraction is to extract a
set of features which maximizes the recognition rate
with the least amount of elements.
Khmer OCR Using Generic Fourier Descriptor Back
Generic Fourier Descriptor
Khmer OCR Using Generic Fourier Descriptor
17
 GFD is derived by applying two-dimensional
Fourier transform on a polar-raster sampled shape
image.
 Polar-raster sampled : the image obtained after
circularly sampling an object in an image up to its
maximum radius.
Example of GFD
18
Khmer OCR Using Generic Fourier Descriptor Back
 The similarity between two shapes are measured
by the city block distance between the two set of
GFDs.
ររkhmer OS UI font and its GFD
Recognition Process
19
Khmer OCR Using Generic Fourier Descriptor Back
 To recognize, we compare the GFD’s score
 The score is obtain by comparing the CCs features
vector with CCs feature in database.
Recognition Process : Score
20
Khmer OCR Using Generic Fourier Descriptor Back
 The similarity between two shapes is measured by
the City-Block distance of the two feature vectors of
the shape.
 The lower value means the more similar the shapes
are.
Generic Fourier Descriptor In detail
Khmer OCR Using Generic Fourier Descriptor
21
 GFD is derived by applying two-dimensional
Fourier transform on a polar-raster sampled shape
image.
 A region-base shape descriptor proposed by
Dengsheng Zhang & Guojun Lu. It is confirm to
outperforms common contour-based and region-based
shape descriptors.
Algorithm to computer GFD
Khmer OCR Using Generic Fourier Descriptor
22
 Input the binary shape image data f(x, y)
 Get centroid of the shape (xc, yc)
 Set the centroid as the origin
 Compute Polar Fourier Transform
 Calculate Fourier Descriptor
 Output GFD
Polar Transform
23
 It is the transforming into polar image f(r, θ), for an
input image f(x, y).
c
c
cc
xx
yy
yyxxr
−
−
=−+−= arctan,)()( 22
θ
∑∑
−
=
−
=
==
1
0
1
0
1
and
1
where
M
y
c
N
x
c y
N
yx
M
x
shapetheof(barycent)centroidofrcoordinatotheis),( cc yx
Khmer OCR Using Generic Fourier Descriptor Back
Polar Raster Grid
24
 Polar raster sampling
Polar image Polar raster sample image in Cartesian space
Khmer OCR Using Generic Fourier Descriptor Back
Polar Raster Grid
25
 Binary polar raster sample shape images
Khmer OCR Using Generic Fourier Descriptor Back
Shape, shape normalization and its polar transform
2D- Fourier Transform
26
 2-D Fourier transform on polar raster sample
image f(r, θ ):
where 0≤r<R and θi
= i(2π/T) (0≤ i<T); 0≤ρ<R,
0≤φ<T. R and T are the radial frequency resolution and
angular frequency resolution respectively.
∑∑ +−=
r i
i
T
i
R
r
jrfPF )]
2
(2exp[),(),( φ
π
ρπθφρ
Khmer OCR Using Generic Fourier Descriptor Back
Normalization
27
 Translation invariant due to using centroid as
origin.
 Scale normalization is archived due to the
proportional equality:
}
|)0,0(|
|),(|
,...,
|)0,0(|
|)0,(|
,...,
|)0,0(|
|),0(|
,...,
|)0,0(|
|)1,0(|
,
|)0,0(|
{
PF
nmPF
PF
mPF
PF
nPF
PF
PF
area
PF
GFD =
Khmer OCR Using Generic Fourier Descriptor Back
Future Work
28
 Integrate GFD into the system
 Exploit the possibility to improve the accuracy by
using Classifier for Training module
Khmer OCR Using Generic Fourier Descriptor Back
Khmer OCR Using Generic Fourier Descriptor Back
Thank for your attention !!!
Reference
30
Khmer OCR Using Generic Fourier Descriptor Back
[1] V. Kruy. Preliminary Experiment on Khmer OCR. Kameyama Laboratory,
Waseda Univerisy, Japan.
[2] Thesis for master degree, Khmer OCR, Vanna Kruy.
[3] D. Zhang and G. Lu. Shape-based image retrieval using generic Fourier descriptor.
Gippsland School of Computing and Information
Technology. Monash University. Churchill, Victoria 3842, Australia.
[4] Thesis for Doctoral Degree, chapter 6: Generic Fourier Descriptor, Dengsheng
Zhang.
[5] J.C.Rupe. Vision-Based Hand Shape Identification for Sign Language
Recognition. Department of Computer Engineering Kate Gleason College of
Engineering Rochester Institute of Technology Rochester, NY.
Reference
31
Khmer OCR Using Generic Fourier Descriptor Back
[6] D. Dimov. A polar-Fourier-Wavelet’s Transform for Effective CBIR. 3rd
ADBIS
workshop on Data mining & Knowledge Discovery
[7] I. Lengieng, K. Sochenda and C. Sokhour. , Khmer OCR for Limon R1 Size 22
Report, PAN Localization Cambodia (PLC) of IDRC.er OCR
[8] A. Averbuch, R.R. Coifmany , D.L. Donohoz M. Eladx M. Israeli. Fast and
Accurate Polar Fourier Transform.
Department of Computer Science, Tel-Aviv University, Tel-Aviv 69978, Israel.
Department of mathematics, Yale University, New Haven CT 06520-8283 USA
Department of Statistics, Stanford University, Stanford 94305-9025 CA. USA.

Khmer ocr using gfd

  • 1.
    Mr. Sovann EN 5th YearEngineering student Dept. Computer Science & Communication Institute of Technology of Cambodia Phnom Penh, Cambodia Khmer OCR Using Generic Fourier Descriptor
  • 2.
    Content Khmer OCR UsingGeneric Fourier Descriptor 2  Introduction  Khmer OCR System  Pre-processing  Segmentation  Feature Extraction  Recognition Process  Generic Fourier Descriptor In detail  Future work
  • 3.
    Introduction  Optical CharacterRecognition (OCR) is the electronic conversion of scanned images into machine- encoded text.  OCR makes it possible to edit the text, to search for a word or a phrase, to display or to print a copy free of scanning artifacts and so on. 3 Khmer OCR Using Generic Fourier Descriptor Back
  • 4.
    Introduction  The ideasstarted prior to World War II. Throughout the years, many reliable commercial and academic prototypes have been developed in many natural languages.  Still, due to the lacked effort in Khmer OCR, there is no reliable Khmer OCR software. 4 Khmer OCR Using Generic Fourier Descriptor Back
  • 5.
    Introduction  The ideasstarted prior to World War II. Throughout the years, many reliable commercial and academic prototypes have been developed in many natural languages.  Still, due to the lacked effort in Khmer OCR, there is no reliable Khmer OCR software. 5 Khmer OCR Using Generic Fourier Descriptor Back
  • 6.
    Introduction  The ideasstarted prior to World War II. Throughout the years, many reliable commercial and academic prototypes have been developed in many natural languages.  Still, due to the lacked effort in Khmer OCR, there is no reliable Khmer OCR software. 6 Khmer OCR Using Generic Fourier Descriptor Back
  • 7.
    Introduction 7 Khmer OCR UsingGeneric Fourier Descriptor Back  The current system is based on Mr. Vanna Kruy’s work for his master degree at Waseda University.  The Objective is to produce a reliable Khmer OCR system which is independent of Font and Size.
  • 8.
    Khmer OCR System KhmerOCR Using Generic Fourier Descriptor  OCR System consist of four majors stages : 8  Pre-processing  Segmentation  Feature Extraction  Post processing
  • 9.
    Pre-processing Khmer OCR UsingGeneric Fourier Descriptor 9  Pre-processing aims to produce data that are easy for the OCR systems to operate accurately.  The main objectives of pre-processing are :  Binarization  Noise removal
  • 10.
    Binarization 10  Image linearization(thresholding) refers to the conversion of a gray-scale image into a binary image. Khmer OCR Using Generic Fourier Descriptor Binarization of Input Image
  • 11.
    Salt-Pepper Noise removal 11 Salt-and-peppernoise is a kind of noise which is usually caused by small unnecessary dots produced by either the scanner or the source document itself. Khmer OCR Using Generic Fourier Descriptor Back Particle removal
  • 12.
    Segmentation 12  Segmentation aimsto produce each component to be recognized by the system.  The process is to separate the text of a page into each separate line, then to separate each line into Vertical Component, and finally produce each independent symbol. Khmer OCR Using Generic Fourier Descriptor Back
  • 13.
    Connected Component Analysis 13 Connected Component is a group of pixels accumulating together to form a shape.  The process is to separate the text of a page into each separate line, then to separate each line into Vertical Component, and finally produce each independent symbol. Khmer OCR Using Generic Fourier Descriptor Back
  • 14.
    14 Example using CCA KhmerOCR Using Generic Fourier Descriptor Back Word Segmentation using CCA
  • 15.
    15 Example using CCA Mainpart : Sup script : Sub script : Ccdown : Khmer OCR Using Generic Fourier Descriptor Back
  • 16.
    Feature extraction 16  Infeature extraction stage, each character is represented as a feature vector which becomes its identity. The major goal of feature extraction is to extract a set of features which maximizes the recognition rate with the least amount of elements. Khmer OCR Using Generic Fourier Descriptor Back
  • 17.
    Generic Fourier Descriptor KhmerOCR Using Generic Fourier Descriptor 17  GFD is derived by applying two-dimensional Fourier transform on a polar-raster sampled shape image.  Polar-raster sampled : the image obtained after circularly sampling an object in an image up to its maximum radius.
  • 18.
    Example of GFD 18 KhmerOCR Using Generic Fourier Descriptor Back  The similarity between two shapes are measured by the city block distance between the two set of GFDs. ររkhmer OS UI font and its GFD
  • 19.
    Recognition Process 19 Khmer OCRUsing Generic Fourier Descriptor Back  To recognize, we compare the GFD’s score  The score is obtain by comparing the CCs features vector with CCs feature in database.
  • 20.
    Recognition Process :Score 20 Khmer OCR Using Generic Fourier Descriptor Back  The similarity between two shapes is measured by the City-Block distance of the two feature vectors of the shape.  The lower value means the more similar the shapes are.
  • 21.
    Generic Fourier DescriptorIn detail Khmer OCR Using Generic Fourier Descriptor 21  GFD is derived by applying two-dimensional Fourier transform on a polar-raster sampled shape image.  A region-base shape descriptor proposed by Dengsheng Zhang & Guojun Lu. It is confirm to outperforms common contour-based and region-based shape descriptors.
  • 22.
    Algorithm to computerGFD Khmer OCR Using Generic Fourier Descriptor 22  Input the binary shape image data f(x, y)  Get centroid of the shape (xc, yc)  Set the centroid as the origin  Compute Polar Fourier Transform  Calculate Fourier Descriptor  Output GFD
  • 23.
    Polar Transform 23  Itis the transforming into polar image f(r, θ), for an input image f(x, y). c c cc xx yy yyxxr − − =−+−= arctan,)()( 22 θ ∑∑ − = − = == 1 0 1 0 1 and 1 where M y c N x c y N yx M x shapetheof(barycent)centroidofrcoordinatotheis),( cc yx Khmer OCR Using Generic Fourier Descriptor Back
  • 24.
    Polar Raster Grid 24 Polar raster sampling Polar image Polar raster sample image in Cartesian space Khmer OCR Using Generic Fourier Descriptor Back
  • 25.
    Polar Raster Grid 25 Binary polar raster sample shape images Khmer OCR Using Generic Fourier Descriptor Back Shape, shape normalization and its polar transform
  • 26.
    2D- Fourier Transform 26 2-D Fourier transform on polar raster sample image f(r, θ ): where 0≤r<R and θi = i(2π/T) (0≤ i<T); 0≤ρ<R, 0≤φ<T. R and T are the radial frequency resolution and angular frequency resolution respectively. ∑∑ +−= r i i T i R r jrfPF )] 2 (2exp[),(),( φ π ρπθφρ Khmer OCR Using Generic Fourier Descriptor Back
  • 27.
    Normalization 27  Translation invariantdue to using centroid as origin.  Scale normalization is archived due to the proportional equality: } |)0,0(| |),(| ,..., |)0,0(| |)0,(| ,..., |)0,0(| |),0(| ,..., |)0,0(| |)1,0(| , |)0,0(| { PF nmPF PF mPF PF nPF PF PF area PF GFD = Khmer OCR Using Generic Fourier Descriptor Back
  • 28.
    Future Work 28  IntegrateGFD into the system  Exploit the possibility to improve the accuracy by using Classifier for Training module Khmer OCR Using Generic Fourier Descriptor Back
  • 29.
    Khmer OCR UsingGeneric Fourier Descriptor Back Thank for your attention !!!
  • 30.
    Reference 30 Khmer OCR UsingGeneric Fourier Descriptor Back [1] V. Kruy. Preliminary Experiment on Khmer OCR. Kameyama Laboratory, Waseda Univerisy, Japan. [2] Thesis for master degree, Khmer OCR, Vanna Kruy. [3] D. Zhang and G. Lu. Shape-based image retrieval using generic Fourier descriptor. Gippsland School of Computing and Information Technology. Monash University. Churchill, Victoria 3842, Australia. [4] Thesis for Doctoral Degree, chapter 6: Generic Fourier Descriptor, Dengsheng Zhang. [5] J.C.Rupe. Vision-Based Hand Shape Identification for Sign Language Recognition. Department of Computer Engineering Kate Gleason College of Engineering Rochester Institute of Technology Rochester, NY.
  • 31.
    Reference 31 Khmer OCR UsingGeneric Fourier Descriptor Back [6] D. Dimov. A polar-Fourier-Wavelet’s Transform for Effective CBIR. 3rd ADBIS workshop on Data mining & Knowledge Discovery [7] I. Lengieng, K. Sochenda and C. Sokhour. , Khmer OCR for Limon R1 Size 22 Report, PAN Localization Cambodia (PLC) of IDRC.er OCR [8] A. Averbuch, R.R. Coifmany , D.L. Donohoz M. Eladx M. Israeli. Fast and Accurate Polar Fourier Transform. Department of Computer Science, Tel-Aviv University, Tel-Aviv 69978, Israel. Department of mathematics, Yale University, New Haven CT 06520-8283 USA Department of Statistics, Stanford University, Stanford 94305-9025 CA. USA.