Content uploaded by Adnan Shaout
Author content
All content in this area was uploaded by Adnan Shaout on Sep 30, 2014
Content may be subject to copyright.
International Journal of Advanced Research in Computer Science and Software Engineering 4 (7), July - 2014, pp.
xx
Off-line Handwriting Arabic Text Recognition: A Survey
Iman Yousif Adnan Shaout
Collage of Computer Science & Information Technology The University of Michigan - Dearborn
Sudan University of Science and technology The Electrical and Computer Engineering Department
Khartoum, Sudan. Dearborn, Michigan
iymo@yahoo.com shaout@umich.edu
Abstract- Automatic recognition of offline Handwriting in general is a difficult task. However, comparing
with other language, due the nature of Arabic scripts, recognition of Arabic scrip is more difficult and
present unique technical challenges. Recently, researcher’s attentions in this area have increased and
different methods have been applied. This paper survey the recent development in the field of the offline
handwritten Arabic word and text recognition and it provides a comprehensive review of these methods in
each stage of the recognition system. The survey provides a comprehensive state of the art of offline
handwriting Arabic Text recognition. It presents a comparison among all existing techniques with respect to
various characteristics such as recognition rate, etc. The paper also presents a new proposal for an off-line
handwriting Arabic text recognition system based on fuzzy decision trees.
Key Words: offline Arabic Handwriting Recognition, OCR, AI, Artificial Neural Networks, Hidden Markov
Models, k-Nearest Neighbors, Fuzzy logic, Fuzzy Decision Tree.
I. INTRODUCTION
Character recognition has been one of the most fascinating and challenging research areas in the field of image
processing and pattern recognition in recent years. Due to the variability in writing style and sizes, recognition of
handwritten scripts is even more challenging than printed scripts. In general, Character recognition can be defined as
the task of transforming text represented in spatial form of graphical marks into its symbolic representation [1] [2].
Character recognition systems are of two types, on-line and off-line systems. The main difference between
them is that in an on-line system the recognition is performed at the time of writing (e.g. tablet, smart phone) while
the off-line handwritten recognition is performed after the writing is completed (e.g. scanned document). Since more
information is available in on-line Handwritten system, usually the recognition in on-line is easier [1] [2]. Off-line
handwritten character recognition system is very important for the creation of electronic libraries, digital copies of
handwritten documents and data entries. It also can provide a solution for many automated processes tasks for huge
amounts of data such as automatic mail sorting, check processing, signature verification, writer identification and
document analysis, etc.
Although handwriting Arabic is cursive script, most of the research in this area handles isolated characters;
some researchers published papers about Arabic character recognition [3-9], some about Arabic-Indian numerals
[10-19] and some included both [20]. Different method approaches have been used. Most of these methods are
based on neural network, hidden Markov model and fuzzy logic.
The first commercial Optical character Recognition system (OCR) for Latin script was launched in middle of
the 1950’s [21]. However, due to the paucity of researches on Arabic OCR at that time, the first published paper was
in 1975 and the first OCR system for Arabic was made available in the 1990s [22]. The availability of powerful
inexpensive CPUs, open databases for Arabic handwritten characters, and the availability of words and text
recognition researches have caused the researcher’s interest to increase in this area. Several researches have been
focused on new techniques and methods that would reduce the processing time while providing higher recognition
accuracy. The first survey on Arabic off-line recognition area was published in 1995 [23]. Few survey papers that
were dedicated to Arabic script recognition have been published [24-29] since 1995. This paper will present, to our
knowledge, the first survey that focuses on handwriting Arabic words and text recognition. The rest of this paper is
organized as follows: section 2 discusses some characteristics of Arabic script; section 3 describes a general model
International Journal of Advanced Research in Computer Science and Software Engineering 4 (7), July - 2014, pp.
xx
of OCR system and presents the recent research in each stage; section 4 presents different databases that are used for
Arabic handwriting recognition system; section 5 presented the recent off-line Arabic word and/or text recognition
systems with tables of comparisons that summarize the features, classifiers, testing data, and the recognition rates;
section 6 proposed a novel off-line Arabic text recognition system based on fuzzy decision tree and finally section 7
concludes the paper.
2.0 ARABIC SCRIPT CHARACTERISTICS
Arabic is a widely used language, not only by Arabs in more than 23 countries in the Middle East and North
Africa but is also spoken as a second language by several Asian countries in which Islam is the principle religion
(e.g. Indonesia). Also there are several languages that have adopted Arabic alphabetic such as Farsi, Urdu, Malay,
and some West African languages such as Hausa [12] [13].
Figure 1: Isolated Arabic Alphabit
Figure 2: Arabic Chracters Formes
Arabic scripts are a cursive-type which are written in horizontal lines from right to left and have 28 letters as
shown in figure 1. However, some additional letters are used when writing foreign words that contain sounds which
do not occur in standard Arabic or when writing other languages using Arabic alphabetic (e.g. in Urdu
language). Most Arabic alphabets change their form depending on the position within the word. Most alphabets
have four different shapes; isolated; at the beginning; at the middle or at the end as see in figure 2. Some of Arabic
alphabet strokes look exactly the same and only differ by having dots (one, two, or three) above or under the letter
International Journal of Advanced Research in Computer Science and Software Engineering 4 (7), July - 2014, pp.
xx
(e.g. ,and ). Additionally some alphabets have secondary character called Hamza (). This secondary
character can be above the main character (like and ) or under the main character (like ). Figure 3 illustrates
some of Arabic script characteristics using a paragraph of Arabic handwritten.
Arabic text has small marks, called diacritics, which are used as vowels that may change the meaning of a
word. For example the words
) (and)
( have the same main characters but are pronounce differently; the first
word )
( is a noun and it means plants and the second word ( (
is a verb which means a plant and the only
difference between them is that the first word has a diacritic called sukkun() above the last character and the
second has a different diacritic called fat-ha (). Figure 4 shows the Arabic diacritics.
The use of ligature in Arabic text is common. A ligature is defined as over-lapping combination between two
characters. This combination is some time optional like in meem-haa () and laam-meem () or not as in laam-
alef that can have two forms depending on the writing style ( ? ,). The existing of the ligature makes the
segmentation process more challenging. A solution to this problem is to consider ligatures as additional classes [30].
Figure 3: Some of Arabic script characteristic
fat-
ha
Dhammah Kasrah Tanween Fat-h
(double Fat-ha)
Tanween dam
(double Dhammah)
Tanween kaser
(double Kasrah)
sukkun Shaddah Maddah
َ ُ ِ ً ٌ ٍ ْ ّ ~
Figure 4: Arabic diacritics
3.0 OPTICAL CHARACTER RECOGNITION (OCR)
The task of recognizing offline characters is called Optical Character Recognition (OCR). This name came
from converting a scanned document of handwritten, typewritten or printed text using optically digitizing device
such as optical scanner or camera into machine-encoded text [31]. The digital image then goes throw five major
stages as shown in figure 5.
Modified
character
The writing line
line
Overlapping
characters
A word with three
sub-words line
Seen with missing
components
Noon with
missing dot
A word with
connected characters
Lam-Meem
Ligature
International Journal of Advanced Research in Computer Science and Software Engineering 4 (7), July - 2014, pp.
xx
Figure 5: A typical OCR system
3.1 Preprocessing Stage
This is the first step in the character recognition system. The aim of this stage is to present an enhanced version
of the original image to be more suitable for the next stage. The image will pass a number of operations like
filtering, binarization, thinning, smoothing, baseline detection, skew and slant detection. Each of these operations
are explained as follows [32] [33]:
•Filtering (Noise Removal (: Remove all the unwanted pixels which do not belong to the word shape.
•Binarization: Conversion of a gray scale image into binary image where a pixel can have only one of two
values 0 or 1.
•Thinning: Convert the text image to some representation which is easier to process. This representation
could be a skeleton which is a one-pixel thick representation showing the centerlines of the word or it could
be a contour that represent the region of the text by describing its contour. The most popular method for
representing the contour is the Freeman chain code [34]. Figure 6 shows an example.
•Smoothing: this operation is used to reduce the noise or to straighten the edges of the characters.
•Base line detection: this is one of the major challenges in Arabic handwriting recognition and it can affect
the efficiency of the features extraction, the segmentation stage and skew normalization. The aim of this
operation is to find the baseline of each word or sub-word and rotate it on its center of gravity so that the
baseline becomes horizontal. For Arabic handwritten recognition several methods have been used for
detecting the base line. Horizontal projection of the word skeleton in one of those methods [35, 36].
Figure 7 shows a sample. Hanene Boukerma and Nadir Farah develop baseline estimation algorithm based
on Sub-Words [37] because Arabic words contain often more than one part of Arabic word (PAW), and
some time those PAWs have different slant angles within the same word.
Figure (7) : Horizontal Projection Method for Detecting Arabic Baseline.
3.2 Segmentation Stage
International Journal of Advanced Research in Computer Science and Software Engineering 4 (7), July - 2014, pp.
xx
After the preprocessing stage, the text image may need to be dividing into objects. It splits the paragraph into
separate lines and then split these lines into words or sub words then to characters or sub characters to be
recognized. Segmentation is considered to be one of the most important and challenging tasks in the OCR system.
The impact of this stage will affect the overall system performance.
The cursive nature of the Arabic script, the overlapping between the characters, different forms of each letter
depending on its position in the word or the writing style and the presence of the secondary characters like dots,
Hamza and diacritics, are all factors that increase the difficulty of this stage [38][39]. During the past few years there
have been promising attempts by researchers to solve this problem; some of this work is summarized and compared
in Table 1.
Dinges et al. [40] presented a novel locale grouping based method for line segmentation of handwritten Arabic
documents. After using Support Vector Machine (SVM) to classify the entire connected component as PAW or
diacritics, they used their developed distance measures to calculate the nearest neighbors between all PAWs. The
next step they used was a graph based grouping algorithm to build all the candidate lines starting from the first PAW
from the right for all lines. Finally, all the unused PAWs are assigned to the nearest line. There method works fine
with documents of different writers and styles even if the text lines have unequal skews or curvatures.
To makes use of the nature of Arabic script, Eraqi and Abdelazeem [41] combined the local writing direction
information and the neighborhood geometric characteristics to propose a new efficient explicit technique for
segmentation of offline Arabic handwriting which segmented the text into basic graphemes. The proposed technique
applies the Douglas-Peucker algorithm on the skeletonized parts of the offline handwriting images. This method has
proved effective; as it obtained 91.27% of the correctly segmented graphemes using 1000 images containing 1402
Arabic handwritten words and 7960 Arabic handwritten graphemes taken from the IFN/ENIT database.
Lawgali et al. [42] exploited the fact that segmentation points, which occur at the end of a character and the
beginning of the next, are usually located in the region surrounding the baseline to present a segmentation algorithm
of Arabic handwritten words. The segmentation algorithm starts with segmenting the word into sub-words and then
the baseline of each sub-word is computed. The algorithm then deletes all the descended sub-words which have a
starting point below the baseline. The vertical projection is used to find the candidate points for the segmentation.
The algorithm has been tested using 800 handwritten Arabic words taken from IFN/ENIT database and has achieved
82.98% character accuracy. However, this algorithm couldn’t segment the alphabets ( , ) into three segments
rather it only segmented them into one.
Al-Khateeb et al. [43] introduced a words segmenting methods for Arabic handwritten text. After extracting the
connected components (CCS) and distances among different components are analyzed then the statistical
distribution of this distance is obtained to determine an optimal threshold for word segmentation. Meanwhile, an
improved projection based method is also employed for baseline detection. The proposed method has been tested
with 200 images from the IFN/ENIT database and has obtained 85% accuracy.
Al Hamad and Abu Zitar [39] presented a three steps segmentation method. The first step is based on feature-
based Arabic Heuristic Segmentor (AHS) to obtain over-segmentation from the thinned words. Step two applied
Neural-based segmentation point on those initial segmentation points to validate them. Finally, the outputs of the
previous networks are then combined to decide whether a particular segmentation point is valid or not. The
segmentation achieved accuracy of 82.98% on 500 words written by ten writers.
Al Hamad [44] proposed fusion equations for improving the segmentation of word image. This method has two
phases. In the first phase the author applied AHS to place the Prospective Segmentation Points (PSP) in the whole
parts of the word image. In the second phase the author applies Neural-based segmentation technique to examine all
PSPs and identify the invalid ones. This method has been implemented and tested on 425 word image from local
benchmark database and has achieved 88.96% accuracy.
Tamen and Drias [45] tried to overcome the over-segmentation problem in the segmentation stage by pasting
the segmented parts to rebuild the whole character form after the rejection or the ambiguousness decision in the
recognition stage. First, they used multilayer perceptrons (MLP) in the recognition process but in order to improve
International Journal of Advanced Research in Computer Science and Software Engineering 4 (7), July - 2014, pp.
xx
the system performance, they replaced the MLP by a linear feed forward network. The training was done using the
back propagation algorithm with all the pre segmented Arabic characters and their different positions written by
three different persons. For testing the system the authors used texts written by three other persons.
Parvez & Mahmoud [46] presented a robust lexicon reduction segmentation algorithm to segment Arabic
words into graphemes. This method is based on the characteristic of Arabic script; which indicates predictable
segmentations of Arabic characters. The authors tested there method on 32,492 images from the IfN/ENIT database.
Osman [38] developed a segmentation algorithm for Arabic handwriting. The first step in the algorithm was to
divide the selected image into lines and sub-words, then trace the sub-word contour. Finally, the algorithm detects
the exact points where the contour changes its state from a horizontal to vertical or curved line and consider those
point as a segmentation points. The algorithm achieved 89.4% segmentation accuracy on 537 tested words from the
IFN/ENIT database.
Samoud et al. [47] presented two combining methods for segmenting Arabic handwritten script into characters.
The first method was based on the analysis of the contour (Min-Max) minima and maxima and the projection. The
second method was based on Hough Transform (HT) and Mathematical Morphology (MM) operators. To compare
the two methods the authors used three evaluation criteria’s; segment positions (SP), segment numbers (SN) and the
recognition rates. For the two methods, the segmentation rate was less than 30% on a data set from the IFN/ENIT
Database.
Table 1: Comparison between some segmentation methods
Author Year Segmentation
Scope
Test Data Segmentation Method Accuracy
Dinges et al. 2013 Line
segmentation
- Grouping based method -
Eraqi and
Abdelazeem
2012 Graphemes 1402 words 91.27%
Lawgali et al. 2011 Characters 800 words Extracting baseline 82.98%
Al-Khateeb et al. 2008 Words 200 images Component-based
method
85%
AlHamad and
Abu Zitar
2010 Characters 500 words Over-segmentation &
ANN
82.98%
Tamen and Drias 2010 Characters Texts written by
three persons
Multilayer perceptrons
then ,back propagation
Unknown
Parvez &
Mahmoud
2013 Graphemes 32,492 images Robust lexicon reduction
Al Hamad 2013 Characters 425 word
images
AHS and Neural 88.96%
Osman 2013 Lines, sub-
words and
characters
537 words Contour extracting points 89.4%
Samoud et al 2012 Characters 1250 images Min-Max-projection
HT-MM
Less than
30%
3.3 Feature Extraction Stage
This is also an important stage in the OCR system. Feature extraction is the process of getting useful
information from the word/character image. The information will be used to generate modules to train the classifier
and to be used for classification purposes [48]. In general there are two categories of features extracted, structural
and statistical features. Choosing the wright feature extraction method might be the most important step for
achieving a high recognition rate [49]. However, in some cases the combination of several features extraction types
could be a wise decision to enhance the overall recognition performance.
International Journal of Advanced Research in Computer Science and Software Engineering 4 (7), July - 2014, pp.
xx
Structural features are the character/word image geometrical and topological information. Those obtained
information include the number of PAWS, descenders, ascenders, dot below the baseline, above the baseline, etc.
Figure 8 shows a structural features example. Statistical feature are numerical measures computed over the images.
They include pixel densities, histograms of chain code directions, moments, Fourier descriptors, etc. [50] [51]. In
HMM classifier based system, it is usual to use sliding windows for extracting features from the word image [52].
Figure 8: Structural Features of Tunisian Town Name !"#$ that contain Two words and seven Paws
3.4 Training and Recognition (Classifications) Stage
This stage is considered as the primary stage for the OCR system. It depends on the previous stages so that
defect in the earlier stages will affects the recognition process and will lead to a low recognition rate. More
information about this stage will be covered in the classifications methodology section.
3.5 Post processing Stage
This is the final stage in the OCR system. This stage can improve recognition accuracy and the system
performance by refining the decisions taken by the previous stage and possibly recognizing words by using the
context [1] [46].
4.0 ARABIC HANDWRITING WORDS AND TEXT DATABASE
With the increasing interest in Arabic handwriting recognition, the need for a freely standard Arabic
handwriting database that represents variety of handwriting styles is highly required. In the past, the lack of freely
available Arabic databases is considered as one of the reasons for the lack of research on Arabic text recognition
compared with other languages. Most of research groups implemented their system on set of data gathered
individually. Therefore, the comparison of OCR Arabic systems was not reasonable in the past. Currently, there are
several Arabic text and words databases to serve handwritten Arabic characters, digit, word and text recognition
research. Table 2 illustrates some of Arabic text and words databases.
Al-Ohali et al. [53] in the Centre for Pattern Recognition and Machine Intelligence (CENPARMI) in Montréal
has developed a Database that can be used by researcher in the field of Arabic handwritten Arabic legal amounts
recognition. What most distinguishes this database is the data are extracted from real-life cheques collected from a
financial institution which make recognition systems more adjustable to real-world applications. The database
contain 29498 samples of Arabic sub-words within the domain of legal amount, 15175 samples for Indian digits and
2499 samples of each of legal and courtesy amounts written in Indian digits. CENPARMI database is divided into
training and testing sets. The training set includes 66–75% of the available data. The division between training and
testing data was done randomly.
International Journal of Advanced Research in Computer Science and Software Engineering 4 (7), July - 2014, pp.
xx
Another database than can be used for legal amounts recognition researches is the AHDB (Arabic Handwritten
Database) that was built by Al-Ma’adeed et al. [54]. This database were written by 100 writers and contains words
and sentences that were used in writing checks. It has the most popular written Arabic words and it contains free
handwriting pages in any area of writer interest.
The IFN/ENIT database of handwritten Tunisian town names [55] is the most commonly used databases by
researchers who are working on Arabic handwritten recognition systems. This database was developed by the
Institute of Communications Technology (IfN) at Technical University Braunschweig in Germany and the Ecole
Nationale d’Inge’nieurs de Tunis (ENIT) in Tunisia. Version 1.0 of the IFN/ENIT-database consists of 26459
handwritten Tunisian town/village names, 115585 pieces of Arabic words (PAWs), and 212211 characters. Each
handwritten town name comes with binary image bitmap and additional GT information. Several competitions in the
past few years have been conducted using this database [56, 57, 58, 58 and 60]. Also, most of the research that was
published recently have used this database.
Mezghani et al. [61] introduced an Arabic handwritten text images database written by multiple writers
(AHTID/MW). The AHTID/MW contains 3710 text lines and 22896 word images written by 53 writers.
A Research group from Sudan University of Science and Technology has developed SUST-ALT database
(Sudan University of Science and Technology- Arabic Language Technology group) [62]. The SUST-ALT database
contains numerals datasets, isolated Arabic letters datasets and Arabic names datasets. Most of these datasets are
off-line.
A research group from King Fahd University of Petroleum & Minerals (KFUPM), Dhahran, Saudi Arabia has
developed Arabic Offline Handwritten Text Database KHATT (KFUPM Handwritten Arabic Text) [63, 64]. The
database were written by 1000 different writers from different countries, gender, age groups, and handedness and
education level. The database contains 2000 similar-text paragraph images and 2000 unique-text paragraph images
and their extracted text line images.
Lawgali et al. [65] has developed a new database for handwritten Arabic characters (HACDB). Although this
is a characters database but it can be used for training and testing words recognition after the segmentation stage
because it cover all shapes of Arabic characters including overlapping ones. The HACDB contains 6600 shapes of
characters written by 50 writers.
Developing a database for offline Arabic handwritten text is expensive in term of manpower and time.
Therefore, Dinges et al. [66] developed an efficient system that automatically generates images of synthetic
handwritten words or text lines from Unicode. The system is based on an Active Shape Models (ASM) that used
online sample to generate unique letter representations for any chosen synthesis. These representations are modified
by affine transformations, smoothed by B-Spline interpolation and composed to text. This system can be used as
alternative to off-line handwritten samples, with variations in shape and texture.
International Journal of Advanced Research in Computer Science and Software Engineering 4 (7), July - 2014, pp.
xx
Table 2: Samurais of Arabic Text and Words Databases
5.0 Classifications Methodology
Compared to Latin and Chinese script where a lot of research work was done, the number of published
work on Arabic script is quite limited. Although the number of papers published in the past few years is increasing
and different techniques and methods which were intended to reduce the processing time while providing higher
recognition accuracy are reported. Most of those classifying methods are based on Artificial Neural Networks
(ANN), Hidden Markov Models (HMM), k-Nearest Neighbors (k-NN), Fuzzy logic (FL), Hybrid approaches and
others. In general there are two basic strategies for recognizing words; Holistic strategies (Global) or Analytic
strategies. The Holistic (Global) strategy recognizes the whole words or sub words without requiring segmentation,
but it works on a limited number of vocabularies. The Analytic Strategy recognizes the segmented features, requires
segmentation, and can be applied on unlimited vocabularies [33]. The rest of this section is categorized according to
these strategies and some related works will be illustrated.
5.1 Holistic Strategies
Literal Amounts recognition is one of the most important applications in offline handwriting recognition area.
Few decades ago processing checks without human involvement was just a dream. Since Literal Amounts contains a
limited lexicon (48 words that can be written in an Arabic literal check
amount
) it seems reasonable that all the
researchers in this area have used the holistic approach.
In recent years, some promising recognition systems using holistic approach have been published. Table 3
shows a summary of those attempts. In 2004 Farah et al. [67] have presented an offline Arabic check literal amount
Database name No of writers Contents
CENPARMI - Database for
handwritten Arabic checks
[53]
Real-life data 2499 words
29498 sub-words
15175 Indian digits
AHDB [54] 100 writers Words and sentences that used in in writing checks.
popular Arabic words
Free handwriting pages.
IFN/ENIT database[55] 411 writers 26459 handwritten Tunisian town/village names
115585 PAWs
212211 characters
AHTID/MW [61] 53 writers 3710 text lines
22896 word images
SUST-ALT database[62] numerals datasets
isolated Arabic
letters datasets
Arabic names datasets
KHATT database[63] [64] 1000 writers 2000 similar-text paragraph images and their extracted
text line images.
2000 unique-text paragraph images and their extracted
text line images.
HACDB[65] 50 writers 6600 segmented characters
International Journal of Advanced Research in Computer Science and Software Engineering 4 (7), July - 2014, pp.
xx
recognition system. The system is based on three parallel classifiers (ANN multilayer perceptron, k-Nearest
Neighbors, and a fuzzy KNN). The input to all the classifiers is the same set of structural features. The results of the
three classifiers are combined using a statistical decision system. The system obtained 96% recognition rates using a
database containing 4800 words which represents the 48 words of lexicon written by 100 different writers of which
1200 wards were used for training and the rest were used for testing. Although the achieved recognition rate is
satisfying, the system is not suitable for a large vocabulary lexicon.
In 2005 Farah with different group [68] has used the same database to present another Arabic literal amount
recognition system. But this time, they used structural and statistical feature extraction and three parallel neural
network classifiers (Multi-Layer Perceptron (MLP)). The obtained results were then combined to produce a final
decision. This time the best recognition rate was 93.00%. In the same year the same group [69] produced another
system that used parallel neural network classifiers feed by structural and the statistical feature extraction, but, this
time one of the MLP was used as a Meta classifier. The same database has been use (2400 words for training and the
2400 for testing). Different parallel combination schemes were presented, and the best recognition rate was 95.2%.
Souici-Meslati & Sellami [70] have presented a hybrid neuro–symbolic classifier approach for recognition
Arabic literal amount. The knowledge base constructed using features extracted from 48 words and they used a
translation algorithm to convert the rules representation into a neural network. This system obtained 93%
recognition rate. In order to evaluate this system, the authors have used 576 words written by four different writers
for training and 1200 words written by 25 different writers for testing.
Based on fuzzy proximity measure especially in bank checks area L Farah et al. [71] presented another literal
amount recognition system using fuzzy classifier to allocate a class to the test word on a basis of a training set. The
fuzzification was introduced in two stages. The first stage was to reclassify the obtained K nearest neighbors (KNN)
by a crisp KNN approach. The second stage in the classification of the tested word was to allocate it to a class
among its K neighbors. The system obtaining of 93.80% recognition rate using 1200 images of 48 words written by
25 different writers.
Automatic recognition of city names and addresses in large quantity of mail is highly essential. Although postal
addresses have a large vocabulary compared to literal amount lexicon. However, Holistic approach is also wildly
used in postal addresses recognition.
Based on decision tree classifier, Amrouch et al. [72] has presented an offline handwritten Algerian city names
recognition system. The authors have used structural features (sub words, ascenders, descenders, loops and
diacritical dots) of the word images as an input for the decision tree. The system achieved 75.74% recognition rate
using database contains 48 city name written three times by 100 writers for learning and testing the system.
Souci et al. [73] have presented an Arabic postal code recognition system. The system is a knowledge based
artificial neural network. The first step in this system is to localize the city name from the envelope and segments it
into words, then structural feature are extracted from the word contour. The knowledge base rule sets were
constructed using a description of the words features. The rules are then translated by spatial algorithm for the neural
network, which is trained in 550 words of 55 Algerian city names written by ten different writers. Using the same
training set, a comparison was carried out by the authors between their proposed system and a MLP classifier
system. The comparison showed that the training took about 10 times less than the MLP classifier and the best
recognition rate achieved was 92%.
Table 3: Summary of results for Literal Amounts and City Name Recognition Systems
authors Yea
r
Representa
tion
Feature Classification
methods
Training data Testing data Recognition
Rates
Farah et al.
[1]
2004 Structural
contour
structura
l
ANN, KNN &
Fuzzy KNN
1200 word images 3600 word
images
96%
International Journal of Advanced Research in Computer Science and Software Engineering 4 (7), July - 2014, pp.
xx
Farah et al.
[2]
2005 contour &
diacritical
dots
structura
l &
statistica
l
Three Multi-Layer
Perceptron (MLP)
ANN
ANN Multi
classifiers
2400 2400 93.00%
Farah et al.
[3]
2005 contour &
diacritical
dots
structura
l &
statistica
l
ANN Multi
classifiers
2400 2400 95.2%
Souici-
Meslati &
Sellami[4]
2004 Contour structura
l
neuro–symbolic
classifier
576 words
written three times
by four writers
1200
48 words
written by 25
93%
L Farah et
al[5]
2006 Contour structura
l
Fuzzy K-NN - 1200 words 93.80%.
Amrouch et
al. [72]
2011 contour structura
l
decision tree 14.400 words 14.400 words 75.74%
Souci et al.
[73]
2004 contour structura
l
knowledge based
artificial neural
network
550 words 550 words 92%.
5.2 Analytic Strategies
This Segmentation base recognition method is suitable for large vocabulary recognition system.
Segmenting the word/sub-word into characters is required. Analytic strategies are divided into two categories,
implicit and explicit base segmentation. In the Implicit based segmentation the segmentation and recognition of
characters are achieved at the same time. The system searches the image for components that match the predefined
classes were in explicit base segmentation the segments are identified based on “character like” properties [74] [75].
5.2.1 Hidden Markov Models Approach
The success of using Hidden Markov Models (HMMs) methods in automatic speech recognition
encouraged researcher to use it in hand written recognition [76]. HMM is considered as one of the most commonly
and successfully used method in offline Arabic handwritten word recognition [52] [76-84].
Based on combined scheme of HMMs and re-ranking, Al Khateeb et al. [85] have presented an Offline
Arabic
handwritten
text
recognition system. The proposed system has three main stages;
preprocessing,
feature extraction and classification. The features were extracted from the segmented words using sliding window.
The extracted features are fed to the HMM classifier. In order to improve accuracy, the HMM result is further
refined by using a re-ranking Scheme. Using the IFN/ENIT database, the system has achieved 95.15% recognition
rate.
Using an explicit segmentation module, Elzobi et al. [86] have presented an off-line Handwriting Arabic
words recognition system based on Hidden Markov Model. Instead of using sliding window based features; they
used shape representative features for each letter in each handwritten form. They have used two databases; the
IESK-arDB for training and testing, and the IFN/ENIT database samples for validation. The recognition rate have
reached 71% on the first database and only 42% for the second (due to the variability in IFN/ENIT which is higher
than that of IESK-arDB).
5.2.2 Artificial Neural Network Approach (ANN)
Neural network approach has performed successfully in many fields and off-line handwritten recognition is
one of them. The ability to be trained automatically from examples, have faster development times , possible run on
parallel processors and achieving good performance with noisy data, makes the use of ANN as a classifier appealing
[87-94].
International Journal of Advanced Research in Computer Science and Software Engineering 4 (7), July - 2014, pp.
xx
Farah [7] has implemented a neural network based system that used cascaded networks to recognize Arabic
segmented characters after resizing the character image to 48 X 32 pixels block, filtration, and converting to binary
in the preprocessing stage. The segmented character is further divided to 6 by 4 segments. Five features are
extracted from each of 24 blocks. Those 120 features of the character are being passed to the neural network input
as a single column. The basic structure of the neural networks consists of one MLP network and three LVQ
networks. The data features of the image is being inputted to the MLP network. In order to minimize the complexity
of the network the similar characters are being recognized as the same ( , , , , , ,% & ' ( ) * +). The output of
first neural network w ill be the inpu t of the LVQ networks after dividing it into three categories. LVQ
networks have the ability to recognize very close features with lower processing time. The data set that was used to
test and measure the proposed system performance consisted of 100 different separated characters that have been
written by 10 different persons with the Arabic characters “Roq’a” style, the most common Arabic writing style. The
recognition rate of this system was between 51% and 77% based on the character shape.
5.2.3 Fuzzy logic Approach
Using fuzzy logic in Arabic handwritten recognition seems to be very logical. The script and the variability of
the Arabic script makes automatic Arabic recognition a very challenging task. A fuzzy set is similar to a classical
set except that in a classical set data can either belong to the set or not whereas in a fuzzy set the data will always
belong to the set but with a different degree. The degree of belonging to a fuzzy set is called a Membership [95]
[96].
In 1994 Abuhaiba et al. [97] have presented an automatic off-line character recognition system for handwritten
cursive Arabic characters. The system is divided into two stages, preprocessing and recognition stage. In the
preprocessing stage, the first step was to skeletonize the segmented character using clustering-based skeletonization
algorithm (CBSA) then the character skeleton is convert to a tree structure for recognition. In the recognition stage,
a set of fuzzy constrained character graph models (FCCGM’s) was designed. For recognition, a set of rules was
applied to match a character tree structure to an FCCGM. The system achieved 73.6% recognition rate with 420
characters used for learning and 330 for testing.
In order to show the importance of the intuitionistic fuzzy similarity measures (IFSM), Baccour et al. [98] have
applied the IFSM on a data set from the IFN/ENIT database. After extracting the features from the word image,
these features were fuzzified and represented by intuitionistic fuzzy sets. IFSMs then were applied to make the
comparison between the test data set which was made of 4357 word images and the training data set which was
made of 2180 word images. The best obtained recognition rate was 90.78%.
Parvez and Mahmoud [20] have presented a novel method for recognizing isolated Arabic handwritten
alphanumeric characters. After the preprocessing stage, the contour of the character image was extracted and
polygonal approximation of the contour was constructed. The nearest neighbor (NN) classifier based on fuzzy
attributed turning function (FATF) was used for classification. For testing and system performance the authors have
used two different databases, one for handwritten Arabic characters and other for Arabic numerals. The system
obtained a recognition rates of around 98% for Arabic characters and more than 97% for Arabic numerals. Then in
2013 the same authors [30] have extend their work to present the first integrated offline Arabic handwritten text
recognition system based on structural techniques. In addition, they introduced several novel ideas and techniques
that can be used for
structural
recognition of Arabic handwriting. The first step in this system was to extract the
PAWs from the text lines, then perform a novel slant corrected algorithm at the PAW level. A novel segmentation
algorithm was then used which was integrated into the recognition phase. The PAW were segmented into smaller
components were these components may be valid Arabic characters or parts of Arabic characters. After that the best
segmentation of the PAW and its constituent characters were identified by an adaptive algorithm. Multiple
hypotheses were also generated for each PAW and passed through post-processing steps, like lexicon consultation,
to re-rank the hypotheses and select the best matching word.
In the training phase, a m
odeling
of
Arabic
iso la ted characters was
done
by
polygonal approximation of the
characters contours.
The
resulting
models, called Fuzzy Attributed Turning Functions
(FATF). T
he authors compared there
system with other systems using the
IfN/ENIT database and achieve 79.58% recognition rate.
International Journal of Advanced Research in Computer Science and Software Engineering 4 (7), July - 2014, pp.
xx
The work in [30] was extended by Mahmoud et al. [64] where they developed an open vocabulary offline
handwritten Arabic text structural recognition system. In the recognition stage, the basic shape of the PAW without
dots did pass through two levels. The first level was used to generate hypotheses for the PAW image then the
segmented part from the PAW was matched with the characters model using a fuzzy distance measure. The second
level generated hypotheses of the paw which was verified to leave only the best hypotheses from the first level.
Finally PAW dot information wear incorporated to generate the final PAW hypotheses. The open vocabulary offline
handwritten Arabic text structural recognition system was tested using 7900 isolated characters written by 52
writers. The system achieved 51.5% recognition rate using KHATT database.
5.2.4 Hybrid approach
Leila et al. [99] have presented an off-line Multiple Classifier System (MCS) to solve Arabic cursive word
recognition problem. This system has two different classifiers, the Fuzzy Adaptive Resonance Theory (Fuzzy ART
network) which was used for the first time in Arabic OCR, and the Radial Basis Functions (RBF). Using IFN/ENIT
database the combined system had a recognition rate of 90.1 %.
Nemouchi, et al. [100] have produced a multi classifiers system for Arabic handwritten words recognition. The
proposed system focused on two phases, the feature extraction and classification phases. In this system the words
were represented using three feature extraction methods. The Zernike moments were extracted from binary image,
the Freeman chain code was extracted from the image contour, and zoning was done on the image skeleton. Those
extracted feature were used as inputs to the four parallel classifiers; Fuzzy C-Means algorithm (FCM), K-Means
algorithm, K Nearest Neighbor algorithm (KNN) and a Probabilistic Neural Network (PNN)). When using all
features the system obtained 80% recognition rate on 1440 words images from the Algerian city-name images
database.
Farsi language uses Arabic alphabetic for writing. Therefore it seems reasonable to mention the researches that
addictive to this language. Based on fuzzy vector quantization (FVQ) and hidden Markov model (HMM) Dehghan
et al. [101] have presented a postal address recognition system. The proposed system was tested using 17,000
images of 198 Farsi city names with the best recognition rate of 96.5%.
6.0 A new Proposed Fuzzy Decision Tree Method
Decision trees are considered a powerful solution structures for many applications like pattern recognition,
machine learning and data mining [102]. They are capable of breaking down complex decisions into simpler
decisions that can be managed making them suitable for classification problems [103] [104]. Decision trees have
been used once for off-line Arabic word handwritten recognition by Amrouch et al. [72]. They were also used for
printed Arabic text recognition by A. Amin [105] and by Abuhaiba [106] for Arabic printed font recognition.
Decision tree was also used for handwritten Urdu and Bangla characters recognition [107] [108].
Recently, fuzzy set theory has been combined with decision trees to produce a powerful tool that can deal with
ambiguity and vagueness in real life problem. This combination is known as fuzzy decision tree which was firstly
introduced by Chang and Pavlidis [109] in 1977. Since then Fuzzy decision trees have played important roles in
many fields such as pattern recognition and classification. Gaolin Fang et al. [110] have used fuzzy decision tree
with heterogeneous classifiers to develop Large Vocabulary Sign Language Recognition system. Kasim et al. [111]
have used fuzzy decision tree to develop image classifier for Batik, one Indonesian cultural heritage image
classification. Decision tree have been used in Speech Recognition [112] and in the medical field for diagnosing
breast cancer [113].
Despite that success in those areas, for unknown reason fuzzy decision trees have never been used for Arabic
handwritten recognition. We think fuzzy decision tree will play a major role and will achieve essential results in
recognizing Arabic handwritten. Therefore, we are planning to develop an off-line Arabic handwritten text
recognition system based on fuzzy decision trees. We are planning to use the IFN/ENIT database of handwritten
Tunisian town names, which is consider to be the most commonly used databases by off-line Arabic recognition
researchers for training and testing.
International Journal of Advanced Research in Computer Science and Software Engineering 4 (7), July - 2014, pp.
xx
7.0 CONCLUSIONS:
This paper provides a comprehensive state of the art presentation of offline handwriting Arabic word and
text recognition. The paper presented the unique characteristics of handwritten Arabic text and word. It also
presented a survey of the recent development in the field of the offline handwritten recognition and provided a
comprehensive review of these methods in each stage of recognition system and surveyed the existing Arabic word
and text database.
Although Arabic language is a cursive written language, yet most of the research in literature was directed
to isolated character recognition and relatively few for word and text recognition. Therefore, it is clear that offline
recognition of Arabic text is still an open issue. There is still urgent need for a high speed recognition rate systems.
The improvements in any stage of recognition system will lead to increasing of the overall system efficiency.
Therefore, more research is needed in all the recognition system stages especially the segmentation and the
classification stages, since they are the most challenging tasks in the offline Arabic handwritten recognition system.
New and improved offline Arabic handwritten recognition systems can be generated through extracting
different kind of features, combining between different technologies or experiencing techniques that have never
been used before. We believe that the use of fuzzy decision trees could lead to a remarkable new offline Arabic
handwritten recognition system.
REFERENCE
[1] Ahmed, Pervez, and Yousef Al-Ohali. "Arabic character recognition: Progress and challenges." Journal of
King Saud University-Computer and Information Sciences 12 (2000): 85-116.
[2] AL-Shatnawi, Atallah Mahmoud, AL-Salaimeh, Safwan, AL-Zawaideh, Farah Hanna, Omar, Khairuddin,
2011. Offline Arabic text recognition an overview. World Computer. Sci. Inform. Technol. J. 1 (5), 184–192
[3] Ali, Mohamed A. "A classifier for Arabic handwritten characters based on supervised self-organizing map
neural network." Proceedings of the 2010 international conference on Mathematical models for engineering
science. World Scientific and Engineering Academy and Society (WSEAS), 2010.
[4] Abed, Majida Ali, H. Abed, Z. Baha, and A. Ismail. "Fuzzy Logic approach to Recognition of Isolated Arabic
Characters." Int. Jour. of Computer Theory and Engineering 2, no. 1 (2010): 119-124.
[5] Elglaly, Yasmine, and Francis Quek. "Isolated Handwritten Arabic Character Recognition using Multilayer
Perceptrons and K Nearest Neighbor Classifiers." (2011).
[6] Gharehchopogh, Farhad Soleimanian, and Ezzat Ahmadzadeh. "Artificial Neural Network Application in
Letters Recognition for Farsi/Arabic Manuscripts." International Journal of Scientific & Technology Research
1.8 (2012): 90-94.
[7] Zawaideh, Farah Hanna. "Arabic Hand Written Character Recognition Using Modified Multi-Neural
Network." Journal of Emerging Trends in Computing and Information Sciences (ISSN 2079-8407) 3.7 (2012):
1021-1026.
[8] Dinges, Laslo, Ayoub Al-Hamadi, and Moftah Elzobi. "An Active Shape Model based approach for Arabic
handwritten character recognition." Signal Processing (ICSP), 2012 IEEE 11th International Conference on.
Vol. 2. IEEE, 2012.
[9] Sahlol, Ahmed, and Cheng Suen. "A Novel Method for the Recognition of Isolated Handwritten Arabic
Characters." arXiv preprint arXiv: 1402.6650 (2014).
[10] Mahmoud, Sabri A., and Sameh M. Awaida. "RECOGNITION OF OFF-LINE HANDWRITTEN ARABIC
(INDIAN) NUMERALS USING MULTI-SCALE FEATURES AND SUPPORT VECTOR MACHINES VS.
HIDDEN MARKOV MODELS." Arabian Journal for Science & Engineering (Springer Science & Business
Media BV) 34 (2009).
[11] Montazer, Gholam Ali, Hamed Qahri Saremi, and Vahid Khatibi. "A neuro-fuzzy inference engine for Farsi
numeral characters recognition." Expert Systems with Applications 37.9 (2010): 6327-6337.
[12] Mahmoud, Sabri A., and Sunday Olusanya Olatunji. "Handwritten Arabic numerals recognition using multi-
span features & Support Vector Machines." Information Sciences Signal Processing and their Applications
(ISSPA), 2010 10th International Conference on. IEEE, 2010.
[13] Mahmoud, Sabri A., and Marwan H. Abu-Amara. "Recognition of handwritten Arabic (Indian) numerals using
Radon-Fourier-based features." Proceedings of the 9th WSEAS International Conference on Signal Processing,
Robotics and Automation, (ISPRA’10), ACM Press, USA. 2010.
[14] Lawal, Isah A., Radwan E. Abdel-Aal, and Sabri A. Mahmoud. "Recognition of handwritten Arabic (Indian)
numerals using freeman's chain codes and abdicative network classifiers." Pattern Recognition (ICPR), 2010
20th International Conference on. IEEE, 2010.
[15] ALI, ABDULBARI AHMED, and RJ RAMTEKE. "FUZZY BASED RECOGNITION OF HANDWRITTEN
ARABIC NUMERALS." International Journal of Machine Intelligence 3.3 (2011).
International Journal of Advanced Research in Computer Science and Software Engineering 4 (7), July - 2014, pp.
xx
[16] Azeem, Sherif Abdel, and Maha El Meseery. "Arabic Handwriting Recognition Using Concavity Features and
Classifier Fusion." Machine Learning and Applications and Workshops (ICMLA), 2011 10th International
Conference on. Vol. 1. IEEE, 2011.
[17] Singh, Pratibha, Ajay Verma, and Narendra S. Chaudhari. "Classification of Hindi numeral using Fuzzy
Zoning and SVM." Advanced computer and communication conference. 2011.
[18] Zaghloul, Rawan I., Dojanah MK Bader Enas, and F. AlRawashdeh. "RECOGNITION OF HINDI (ARABIC)
HANDWRITTEN NUMERALS." American Journal of Engineering and Applied Sciences 5.2 (2012): 132.
[19] AlKhateeb, Jawad H., and Marwan Alseid. "DBN-Based learning for Arabic handwritten digit recognition
using DCT features." Computer Science and Information Technology (CSIT), 2014 6th International
Conference on. IEEE, 2014.
[20] M. T. Parvez and S. Mahmoud, “Arabic Handwritten Alphanumeric Character Recognition using Fuzzy
Attributed Turning Functions,” in First International Workshop on Frontiers in Arabic Handwriting
Recognition, 2011.
[21] Eikvil, Line. "Optical Character Recognition." citeseer. ist. psu. Edu/142042. Html (1993).
[22] Märgner, Volker, and Haikal El Abed. "Arabic word and text recognition-current developments." Proceedings
of the Second International Conference on Arabic Language Resources and Tools, Cairo, Egypt, April. The
MEDAR Consortium. 2009.
[23] Al-Badr, Badr, and Sabri A. Mahmoud. "Survey and bibliography of Arabic optical text recognition." Signal
processing 41.1 (1995): 49-77.
[24] Amin, Adnan. "Off line Arabic character recognition: a survey." Document Analysis and Recognition, 1997.
Proceedings of the Fourth International Conference on. Vol. 2. IEEE, 1997.
[25] Amin, Adnan. "Off-line Arabic character recognition: the state of the art." Pattern recognition 31.5 (1998):
517-530.
[26] Khorsheed, Mohammad S. "Off-line Arabic character recognition–a review." Pattern analysis & applications
5.1 (2002): 31-45.
[27] Lorigo, Liana M., and Venugopal Govindaraju. "Offline Arabic handwriting recognition: a survey." Pattern
Analysis and Machine Intelligence, IEEE Transactions on 28.5 (2006): 712-724.
[28] Jumari, Kasmiran, and Mohamed A Ali. "A survey and comparative evaluation of selected off-line Arabic
handwritten character recognition systems." Jurnal Teknologi 36.1 (2012): 1-18.
[29] Parvez, Mohammad Tanvir, and Sabri A. Mahmoud. "Offline Arabic handwritten text recognition: a survey."
ACM Computing Surveys (CSUR) 45.2 (2013): 23.
[30] Tanvir Parvez, M., & Mahmoud, S. a. (2013). Arabic handwriting recognition using structural and syntactic
pattern attributes. Pattern Recognition, 46(1), 141–154. doi:10.1016/j.patcog.2012.07.012
[31] RAJPUT, MR NARENDRASING B., SM RAJPUT, and SM BADAVE. "Handwritten Character Recognition-
A Review." International Journal of Engineering Research & Technology (IJERT) Vol. 1 Issue 8, October –
2012 ISSN: 2278-0181
[32] Suliman, Azizah, Mohd Nasir Sulaiman, Mohamed Othman, and Rahmita Wirza. "Chain Coding and Pre
Processing Stages of Handwritten Character Image File." electronic Journal of Computer Science and
Information Technology 2, no. 1 (2011).
[33] Arica, Nafiz, and Fatos T. Yarman-Vural. "An overview of character recognition focused on off-line
handwriting." Systems, Man, and Cybernetics, Part C: Applications and Reviews, IEEE Transactions on 31.2
(2001): 216-233.
[34] Al-rashaideh, H. (2006). Preprocessing phase for Arabic Word Handwritten Recognition. Information
Transmissions in Computer Networks Journal, 6(1), 11–19. Retrieved from
http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.95.5611
[35] Srihari, Sargur N., and Gregory Ball. "An assessment of Arabic handwriting recognition technology." Guide to
OCR for Arabic Scripts. Springer London, 2012. 3-34.
[36] Atallah, AL-Shatnawi, and Khairuddin Omar. "Methods of Arabic language baseline detection–The state of
art." IJCSNS 8.10 (2008): 137.
[37] Boukerma, Hanene, and Nadir Farah. "A novel Arabic baseline estimation algorithm based on sub-words
treatment." Frontiers in Handwriting Recognition (ICFHR), 2010 International Conference on. IEEE, 2010.
[38] Osman, Yusra. "Segmentation algorithm for Arabic handwritten text based on contour analysis." Computing,
Electrical and Electronics Engineering (ICCEEE), 2013 International Conference on. IEEE, 2013.
[39] Al Hamad, Husam A., and Raed Abu Zitar. "Development of an efficient neural-based segmentation technique
for Arabic handwriting recognition." Pattern Recognition 43.8 (2010): 2773-2798.
[40] Dinges, Laslo, Ayoub Al-Hamadi, and Moftah Elzobi. "A Locale Group Based Line Segmentation Approach
for Non Uniform Skewed and Curved Arabic Handwritings." Document Analysis and Recognition (ICDAR),
2013 12th International Conference on. IEEE, 2013.
[41] Eraqi, Hesham M., and Sherif Abdelazeem. "A new Efficient Graphemes Segmentation Technique for Offline
Arabic Handwriting." Frontiers in Handwriting Recognition (ICFHR), 2012 International Conference on.
IEEE, 2012.
International Journal of Advanced Research in Computer Science and Software Engineering 4 (7), July - 2014, pp.
xx
[42] Lawgali, A., Bouridane, A., Angelova, M., & Ghassemlooy, Z. (2011, September). Automatic segmentation for
Arabic characters in handwriting documents. In Image Processing (ICIP), 2011 18th IEEE International
Conference on (pp. 3529-3532). IEEE.
[43] AlKhateeb, Jawad H., Jianmin Jiang, Jinchang Ren, and S. Ipson. "Component-based segmentation of words
from handwritten Arabic text." International Journal of Computer Systems Science and Engineering 5, no. 1
(2009).
[44] Al Hamad, Husam A. "Neural-Based Segmentation Technique for Arabic Handwriting Scripts." 21st
International Conference on Computer Graphics, Visualization and Computer Vision, WSCG (2013).
[45] Tamen, Zahia, and Habiba Drias. "How to overcome some segmentation problems in a constrained handwritten
Arabic character recognition system." Information Sciences Signal Processing and their Applications (ISSPA),
2010 10th International Conference on. IEEE, 2010.
[46] Parvez, Mohammad Tanvir, and Sabri A. Mahmoud. "Lexicon Reduction Using Segment Descriptors for
Arabic Handwriting Recognition." Document Analysis and Recognition (ICDAR), 2013 12th International
Conference on. IEEE, 2013.
[47] Samoud, Fadoua Bouafif, Samia Snoussi Maddouri, and Hamid Amiri. "Three Evaluation Criteria's towards a
Comparison of Two Characters Segmentation Methods for Handwritten Arabic Script." Frontiers in
Handwriting Recognition (ICFHR), 2012 International Conference on. IEEE, 2012.
[48] Haraty, Ramzi A., and Catherine Ghaddar. "Arabic text recognition." Int. Arab J. Inf. Technol. 1, no. 2 (2004):
156-163.
[49] Lawgali, A., & Bouridane, A. (2011). Handwritten Arabic Character Recognition: Which feature extraction
method. International Journal of Advanced Science and Technology, 34(September)
[50] Naz, S., Hayat, K., Imran Razzak, M., Waqas Anwar, M., Madani, S. a., & Khan, S. U. (2014). The optical
character recognition of Urdu-like cursive scripts. Pattern Recognition, 47(3), 1229–1248.
doi:10.1016/j.patcog.2013.09.037
[51] Elbaati, Abdelkarim, Houcine Boubaker, Monji Kherallah, Abdel Ennaji, and A. M. Alimi. "Arabic
handwriting recognition using restored stroke chronology." In Document Analysis and Recognition, 2009.
ICDAR'09. 10th International Conference on, pp. 411-415. IEEE, 2009.
[52] Märgner, V., El, H., & Mario, A. Offline Handwritten Arabic Word Recognition Using HMM - a Character
Based Approach without Explicit Segmentation. The institute for Communications Technology (IfN),
Technical University of Braunschweig; Department of Signal Processing for Mobile Information Systems
Schleinitzstrasse 22, 38106, Braunschweig, Germany.
[53] Al-Ohali, Yousef, Mohamed Cheriet, and Ching Suen. "Databases for recognition of handwritten Arabic
checks." Pattern Recognition 36.1 (2003): 111-121.
[54] Al-Ma'adeed, Somaya, Dave Elliman, and Colin A. Higgins. "A data base for Arabic handwritten text
recognition research." Frontiers in Handwriting Recognition, 2002. Proceedings. Eighth International
Workshop on. IEEE, 2002.
[55] El Abed, Haikal, and V. Margner. "The IFN/ENIT-database-a tool to develop Arabic handwriting recognition
systems." Signal Processing and Its Applications, 2007. ISSPA 2007. 9th International Symposium on. IEEE,
2007.
[56] V. Märgner, M. Pechwitz, and H. El Abed, “ICDAR 2005 Arabic handwriting recognition competition,” in
Proceedings of the 8th Inter. Conf. on Document Analysis and Recognition, vol. 1, pp. 70–74, 2005
[57] V. Märgner and H. El Abed, “ICDAR 2007 – Arabic Hand-writing Recognition Competition,” in Proceedings
of the 9 the International Conf. on Document Analysis and Recognition (ICDAR), vol. 2, 2007. Märgner, V., &
Abed, H. El. (2009). ICDAR 2009 Arabic Handwriting Recognition Competition. 2009 10th International
Conference on Document Analysis and Recognition, 1383–1387. doi:10.1109/ICDAR.2009.256
[58] Margner, Volker, and Haikal El Abed. "ICFHR 2010-Arabic handwriting recognition competition." Frontiers in
Handwriting Recognition (ICFHR), 2010 International Conference on. IEEE, 2010.
[59] Margner, V., & Abed, H. El. (2011). ICDAR 2011 - Arabic Handwriting Recognition Competition. 2011
International Conference on Document Analysis and Recognition, 1444–1448. doi:10.1109/ICDAR.2011.287
[60] Mezghani, Anis, Slim Kanoun, Maher Khemakhem, and Haikal El Abed. "A database for Arabic handwritten
text image recognition and writer identification." In Proceedings of the 2012 International Conference on
Frontiers in Handwriting Recognition, pp. 399-402. IEEE Computer Society, 2012
[61] Musa, M.E.M., "Arabic handwritten datasets for pattern recognition and machine learning," Application of
Information and Communication Technologies (AICT), 2011 5th International Conference on , vol., no., pp.1,3,
12-14 Oct. 2011.
International Journal of Advanced Research in Computer Science and Software Engineering 4 (7), July - 2014, pp.
xx
[62] Mahmoud, Sabri A., Irfan Ahmad, Mohammad Alshayeb, Wasfi G. Al-Khatib, Mohammad Tanvir Parvez,
Gernot A. Fink, Volker Margner, and Haikal El Abed. "KHATT: Arabic offline handwritten text database." In
Proceedings of the 2012 International Conference on Frontiers in Handwriting Recognition, pp. 449-454. IEEE
Computer Society, 2012.
[63] Mahmoud, Sabri A., Irfan Ahmad, Wasfi G. Al-Khatib, Mohammad Alshayeb, Mohammad Tanvir Parvez,
Volker Märgner, and Gernot A. Fink. "KHATT: An open Arabic offline handwritten text database." Pattern
Recognition 47, no. 3 (2014): 1096-1112.
[64] Lawgali, A., M. Angelova, and A. Bouridane. "HACDB: Handwritten Arabic characters database for automatic
character recognition." Visual Information Processing (EUVIP), 2013 4th European Workshop on. IEEE, 2013.
[65] Dinges, Laslo, Ayoub Al-Hamadi, and Moftah Elzobi. "An Approach for Arabic Handwriting Synthesis Based
on Active Shape Models." Document Analysis and Recognition (ICDAR), 2013 12th International Conference
on. IEEE, 2013.
[66] Farah, Nadir, Labiba Souici, Lotfi Farah, and Mokhtar Sellami. "Arabic words recognition with classifiers
combination: An application to literal amounts." In Artificial Intelligence: Methodology, Systems, and
Applications, pp. 420-429. Springer Berlin Heidelberg, 2004.
[67] Farah, Nadir, Tarek Khadir, and Mokhtar Sellami. "Artificial neural network fusion: Application to Arabic
words recognition." In ESANN, pp. 151-156. 2005.
[68]
[69] Nadir, F. A. R. A. H., E. N. N. A. J. I. Abdelatif, K. H. A. D. I. R. Tarek, and S. E. L. L. A. M. I. Mokhtar.
"Benefit of multi-classifier systems for Arabic handwritten words recognition." In Document Analysis and
Recognition, 2005. Proceedings. Eighth International Conference on, pp. 222-226. IEEE, 2005.
[70] Souici-Meslati, Labiba, and Mokhtar Sellami. "A HYBRID APPROACH FOR ARABIC LITERAL
AMOUNTS RECOGNITION." Arabian Journal for Science & Engineering (Springer Science & Business
Media BV) 29 (2004).
[71] Lotfi, Farah, Farah Nadir, and Bedda Mouldi. "Arabic Words Recognition by Fuzzy Classifier." Journal of
Applied Sciences 6 (2006): 647-650.
[72] Amrouch, Siham, Aida Chefrour, and Labiba Souici-Meslati. "DECISION TREES FOR HANDWRITTEN
ARABIC WORDS RECOGNITION." Proceedings of the International Arab Conference on Information
Technology (ACIT), Erriad-Saudi Arabia. 2011.
[73] Souici, Labiba, Nadir Farah, Toufik Sari, and Mokhtar Sellami. "Rule based neural networks construction for
handwritten Arabic city-names recognition." In Artificial Intelligence: Methodology, Systems, and
Applications, pp. 331-340. Springer Berlin Heidelberg, 2004.
[74] Rehman, Amjad, Dzulkifli Mohamad, and Ghazali Sulong. "Implicit vs explicit based script segmentation and
recognition: a performance comparison on benchmark database." Int. J. Open Problems Compt. Math 2, no. 3
(2009): 352-364.
[75] Choudhary, Amit. "A Review of Various Character Segmentation Techniques for Cursive Handwritten Words
Recognition." International Journal of Information & Computation Technology (IJICT), Volume 4, Number 6
spl. (2014)
[76] Koerich, Alessandro L., Robert Sabourin, and Ching Y. Suen. "Large vocabulary off-line handwriting
recognition: A survey." Pattern Analysis & Applications 6, no. 2 (2003): 97-121.
[77] Kundu, Amlan, Tom Hines, Jon Phillips, Benjamin D. Huyck, and Linda C. Van Guilder. "Arabic handwriting
recognition using variable duration HMM." In Document Analysis and Recognition, 2007. ICDAR 2007. Ninth
International Conference on, vol. 2, pp. 644-648. IEEE, 2007.
[78] Al-Hajj, Ramy, Chafic Mokbel, and Laurence Likforman-Sulem. "Combination of HMM-based classifiers for
the recognition of Arabic handwritten words." In Document Analysis and Recognition, 2007. ICDAR 2007.
Ninth International Conference on, vol. 2, pp. 959-963. IEEE, 2007.
[79] Saleem, S., Cao, H., Subramanian, K., Kamali, M., Prasad, R., & Natarajan, P. (2009). Improvements in
BBN’s HMM-based offline Arabic handwriting recognition system. In Proceedings of the International
Conference on Document Analysis and Recognition, ICDAR (pp. 773–777).
[80] Eraqi, H. M., & Abdelazeem, S. (2012). HMM-based Offline Arabic Handwriting Recognition: Using New
Feature Extraction and Lexicon Ranking Techniques. 2012 International Conference on Frontiers in
Handwriting Recognition, 554–559. doi:10.1109/ICFHR.2012.214
[81] Khorsheed, Mohammad S. "Recognizing handwritten Arabic manuscripts using a single hidden Markov
model." Pattern Recognition Letters 24, no. 14 (2003): 2235-2242.
International Journal of Advanced Research in Computer Science and Software Engineering 4 (7), July - 2014, pp.
xx
[82] Dehghan, Mehdi, Karim Faez, Majid Ahmadi, and Malayappan Shridhar. "Handwritten Farsi (Arabic) word
recognition: a holistic approach using discrete HMM." Pattern Recognition 34, no. 5 (2001): 1057-1065.
[83] El-Hajj, Ramy, Laurence Likforman-Sulem, and Chafic Mokbel. "Arabic handwriting recognition using
baseline dependant features and hidden markov modeling." In Document Analysis and Recognition, 2005.
Proceedings. Eighth International Conference on, pp. 893-897. IEEE, 2005.
[84] Alma'adeed, Somaya, Colin Higgens, and Dave Elliman. "Recognition of off-line handwritten Arabic words
using hidden Markov model approach." In Pattern Recognition, 2002. Proceedings. 16th International
Conference on, vol. 3, pp. 481-484. IEEE, 2002.
[85] Elzobi, Moftah, Ayoub Al-Hamadi, Laslo Dings, Mahmoud Elmezain, and Anwar Saeed. "A Hidden Markov
Model-Based Approach with an Adaptive Threshold Model for Off-Line Arabic Handwriting Recognition." In
Document Analysis and Recognition (ICDAR), 2013 12th International Conference on, pp. 945-949. IEEE,
2013.
[86] AlKhateeb, Jawad H., Jinchang Ren, Jianmin Jiang, and Husni Al-Muhtaseb. "Offline handwritten Arabic
cursive text recognition using Hidden Markov Models and re-ranking." Pattern Recognition Letters 32, no. 8
(2011): 1081-1088.
[87] Lajish, V. L. "A Quick Review of Recognition Strategies Based on Neural Network and Neuro-Fuzzy
Approaches with Special Reference to HCR in Indian Languages." (2013).
[88] Asiri, A., and Mohammad S. Khorsheed. "Automatic Processing of Handwritten Arabic Forms using Neural
Networks." In IEC (Prague), pp. 313-317. 2005.
[89] Zaidan, A. A., B. B. Zaidan, Hamid Jalab, Hamdan Alanazi, and Rami Alnaqeib. "Offline Arabic Handwriting
Recognition Using Artificial Neural Network." arXiv preprint arXiv: 1006.2809 (2010).
[90] Alma'adeed, Somaya. "Recognition of off-line handwritten Arabic words using neural network." In Geometric
Modeling and Imaging--New Trends, 2006, pp. 141-144. IEEE, 1993.
[91] Mohammed, Naji F., and Nazlia Omar. "Arabic named entity recognition using artificial neural network."
Journal of Computer Science 8, no. 8 (2012): 1285.
[92] Perwej, Yusuf. "Recurrent Neural Network Method in Arabic Words Recognition System." arXiv preprint
arXiv: 1301.4662 (2013).
[93] Cheikh, I. Ben, Belaïd, A., Kacem, A., Esstt, U., Hussein, A. T., Mnara, B. P., & Nancy, V. L. (2008). A
Novel Approach for the Recognition of a Wide Arabic Handwritten Word Lexicon Neural model : How to,
benefit from the, 2–5.
[94] Al Hamad, Husam Ahmed. "Use an efficient neural network to improve the Arabic handwriting recognition."
In Signal and Image Processing Applications (ICSIPA), 2013 IEEE International Conference on, pp. 269-274.
IEEE, 2013.
[95] Zadeh L. A., “Fuzzy sets,” Information and Control, Vol. 8, pp. 338-353, 1965.
[96] Zadeh L. A., “Fuzzy Algorithms,” Information and Control, Vol. 12, pp. 94-102, 1968.
[97] Abuhaiba, Ibrahim SI, Sabri A. Mahmoud, and Roger J. Green. "Recognition of handwritten cursive Arabic
characters." Pattern Analysis and Machine Intelligence, IEEE Transactions on 16.6 (1994): 664-672.
[98] Baccour, Leila, and Adel M. Alimi. "A comparison of some intuitionistic fuzzy similarity measures applied to
handwritten Arabic sentences recognition." Fuzzy Systems, 2009. FUZZ-IEEE 2009. IEEE International
Conference on. IEEE, 2009.
[99] Leila, Chergui, Kef Maamar, and Chikhi Salim. "Combining neural networks for Arabic handwriting
recognition." Programming and Systems (ISPS), 2011 10th International Symposium on. IEEE, 2011.
[100] Nemouchi, Soulef, Labiba Souici Meslati, and Nadir Farah. "Classifiers combination for Arabic words
recognition: application to handwritten Algerian city names." Image and Signal Processing. Springer Berlin
Heidelberg, 2012. 562-570.
[101] Dehghan, Mehdi, Karim Faez, Majid Ahmadi, and Malayappan Shridhar. "Unconstrained Farsi handwritten
word recognition using fuzzy vector quantization and hidden Markov models." Pattern Recognition Letters 22,
no. 2 (2001): 209-214.
[102] Pach, Ferenc Peter, Janos Abonyi, Sandor Nemeth, and Peter Arva. "Supervised clustering and fuzzy
decision tree induction for the identification of compact classifiers." In 5th International Symposium of
Hungarian Researchers on Computational Intelligence, Budapest, Hungary. 2004.
[103] MATIAŠKO, Karol, Ján BOHÁČIK, Vitaly LEVASHENKO, and Štefan KOVALÍK. "Learning fuzzy
rules from fuzzy decision trees." Journal of Information, Control and Management Systems 4, no. 2 (2006).
[104] Mitra, S., Member, S., Konwar, K. M., & Pal, S. K. (2002). Fuzzy Decision Tree, Linguistic Rules and
Fuzzy Knowledge-Based Network : Generation and Evaluation, ,32(4), 328–339.
International Journal of Advanced Research in Computer Science and Software Engineering 4 (7), July - 2014, pp.
xx
[105] A. Amin, “Recognition of printed Arabic text based on global features and decision tree learning
techniques”, Pattern recognition, Vol. 33, No. 8, pp. 1309-1323, 2000.
[106] I. S. I. Abuhaiba, "Arabic font recognition using decision trees built from Common Words", Journal of
Computing and Information Technology: CIT, Vol. 13, Num. 3, pp. 211-223, 2005
[107] D.S. Guru, S.K. Ahmed, K. Irfan, An attempt towards recognition of hand-written Urdu characters: a
decision tree approach, in: Proceedings of the National Conference on Computers and Information Technology
(NCCIT'01), 2001, pp. 75–83.
[108] Chowdhury Mofizur Rahman and Monzur Morshed “ Decision tree based learning of handwritten Bangla
characters, ICCIT 99, SUST, Sylhet, December 3- 5,1999.
[109] Chang, Robin L P; Pavlidis, Theodosios, "Fuzzy Decision Tree Algorithms," Systems, Man and
Cybernetics, IEEE Transactions on , vol.7, no.1, pp.28,35, Jan. 1977
doi: 10.1109/TSMC.1977.4309586 decision tree
[110] Fang, G., Gao, W., & Zhao, D. (2004). Large Vocabulary Sign Language Recognition Based on Fuzzy
Decision Trees. IEEE Transactions on Systems, Man, and Cybernetics - Part A: Systems and Humans, 34(3),
305–314. doi:10.1109/TSMCA.2004.824852
[111] Rangkuti, A. Haris, Nashrul Hakiem, Rizal Broer Bahaweres, Agus Harjoko, and Agfianto Eko Putro.
"Analysis of image similarity with CBIR concept using wavelet transform and threshold algorithm." In
Computers & Informatics (ISCI), 2013 IEEE Symposium on, pp. 122-127. IEEE, 2013.
[112] Yang, Shiueng-Bien, and Tzu-Wei Chen. "Fuzzy variable-branch decision tree for speech recognition." In
Asian Control Conference, 2009. ASCC 2009. 7th, pp. 773-778. IEEE, 2009.
[113] Tabakov, Martin, M. Podhorska-Okolow, S. Zareba, and Bartosz Pula. "Using fuzzy Sugeno integral as an
aggregation operator of ensemble of fuzzy decision trees in the recognition of HER2 breast cancer
histopathology images." In Computer Medical Applications (ICCMA), 2013 International Conference on, pp.
1-6. IEEE, 2013.
