CHAKEL-DB: Online Database for Handwriting Diacritic Arabic

Character

Houda Nakkach

, Sofiene Haboubi

and Hamid Amiri

LR-11-ES17 Signal, Images and Information Technology (LR-SITI-ENIT), Tunis El Manar University,

National Engineering School of Tunis 1002, Tunis Le Belvédère, Tunisia

Keywords: CHAKEL-DB, Handwritten Arabic Character, Diacritical Marks, Online Character Recognition.

Abstract: Our paper presents an online database for handwriting Arabic characters, so-called “CHAKEL-DB”: with

“chakel” means diacritic in Arabic language. The database contains 3150 collected samples that present alpha-

numeral characters with diacritical marks. The data are collected from more than 68 writers having different

origins, genders and ages. The data are available in character and stroke levels. We built an elementary

recognition system to test our database and to manipulate a large vocabulary and a huge quantity of variation

style in the collected data. “CHAKEL-DB” is available for the purpose to improve handwriting research field,

to facilitate experiments and researches. This database offers files in XML format.

1 INTRODUCTION

With the increasing spread of hand-held devices such

as PDA, tablet-PC and Smart-phone, the handwriting

is introduced as a new human-computer interaction

modality. As result, an increased demand for a high

performance of the on-line handwritten recognition

system, which becomes a popular and fascinate field

for researcher in the recent years. On-line handwritten

recognition is presented by a sequence of strokes in

2D (x, y) coordinates form, which are the result of

succession points obtained from the pen-down/pen-

up signals. Such data is known as digital ink and can

be considered as a dynamic representation of

handwriting. An on-line handwritten recognition

system typically includes three main elements: An

output display adjacent to a touch sensitive surface, a

database of characters or general gestures and a

recognizer system as a software application which

interprets the movements of the stylus across the

writing surface and translating the resulting strokes

into digital character based on database samples.

In this context, more literature works have been

developed for the Latin characters case, but few of

them for the Arabic case, challenged by the various

nature and unconstrained cursive Arabic Script

(Nakkach et al., 2016) (Biadsy et al., 2011).

https://orcid.org/0000-0003-1699-3812

https://orcid.org/0000-0001-5270-3830

Diacritical marks are marking which are written

either above or below a letter. The diacritics “Fatha”,

“Damma”, and “Kasra” indicate short vowels.

“Sukun” mark indicates a syllable stop, and

“Fathatan” indicates nunation (“tanwin” or double

vowels) that can accompany “Fatha”, “Damma”, or

“Kasra”. The three vowel letters, which are “Alef”,

“Waw” and “Ya” are used to indicate long vowels.

The “chadda” vowel represents doubling (or

gemination) of a consonant (Table 1). So, along with

the dots and other marks representing vowels, this

makes the effective size of the Arabic alphabet about

2040 characters (equation 1). In fact, we have 23

characters with 4 forms, 5 characters with only 2

forms and we have 20-mark models for each form of

character.

(23*4+5*2) *20=2040 (1)

Table 1: Exhaustive list of "BA" character diacritical

models.

Nakkach, H., Haboubi, S. and Amiri, H.

CHAKEL-DB: Online Database for Handwriting Diacritic Arabic Character.

DOI: 10.5220/0008804307430750

In Proceedings of the 15th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2020) - Volume 5: VISAPP, pages

743-750

ISBN: 978-989-758-402-2; ISSN: 2184-4321

743

In addition to these usual issues for the

handwriting recognition, recognizing Arabic

character needs to deal with ligatures, diacritics,

multi-variability writer style, bad writing habits like

touching characters, misplacement of dots, etc.

Therefore, most research in the character recognition

field is moving towards experimentation of their

proposed system to test new approaches and

algorithms. In this case, they need database to

validate their theories, for that they use databases

available for on-line and off-line writing or they

develop their own database as presented in

(Plamondon et al., 2000) (Tagougui et al., 2013)

(Vinciarelli, 2002)( Jäger et al., 2001)(Steinherz et

al., 1999).

On other wise, examining the existing databases,

we find that none of them deals with on-line / off-line

cursive Arabic writing with diacritical marks. Thus,

the development of a new database is useful for the

scientific community, with the continuous grow up of

interested searchers number work on the recognition

of Arabic writing field. To respond to this need, we

propose in this paper a new database for the Arabic

diacritical characters named "CHAKEL-DB".

So, the remainder of this article is organized as

follows: in section 2., we begin with an overview of

the related works. Section 3. presents the data

preparation and acquisition stage. Supplemented by,

a quick overview of the existing data formats and

databases, presented as a comparative study based on

indicators reflecting the requirements to choose the

formats to be adopted in our work. The experimental

results of our work are presented in section 4. And we

close the paper with a conclusion in section 5.

2 RELATED WORK

All scientific domains have several standard

databases for developing, evaluating and comparing

different techniques developed for their various tasks.

The field of recognition of handwriting is not an

exception; in this context, the handwriting

recognition community has proposed many

databases, some of them will be listed below, to

present some statistics and comparisons based on a

few specific dimensions of Handwriting. These

dimensions allow us to make a subdivision in the field

of handwriting. We can divide handwriting

recognition into two fields, depending on the form in

which the data is represented on-line or off-line.

In the case of on-line recognition of handwriting,

the user must write on a digitizing tablet with a special

stylet, so that the lines are sampled by the coordinates

(x, y) of the spaced time intervals. However, in the

case of off-line handwriting recognition, the user

writes on a paper which is then scanned by a scanner.

In this case the data is presented to the system as an

image, which requires segmentation to binarize it

through the threshold technique based on the color

pattern (color or gray scale), so, the image pixels are

either 1 or 0.

The on-line case concerns a spatial-temporal

representation of the input; while the off-line case,

involves an analysis of the spatial luminance of

image.

The most important and widely used handwritten

databases include:

 IAM databases: used as collections of handwritten

samples; they are adopted for a variety of

segmentation and recognition tasks. Several off-

line and on-line databases have been developed

within the IAM, such as:

 IAM-DB (IAM handwriting database) (Marti and

Bunke, 1999): this handwritten database proposed

since 1999, contains forms of unconstrained

western handwritten English text. The IAM

Handwriting Database 3.0 published in 2002

includes contributions from 657 authors making a

total of 1539 handwritten pages including 5685

sentences, 13353 lines of text and 115 320 words.

The database is labelled at the level of sentences,

lines and words, and it has been widely used in

word tracking, writer identification, text

segmentation Handwriting and off-line write

recognition. This database is presented by image

files described by meta-data files in XML format.

 IAM-OnDB (Liwicki and Bunke, 2005): The

database is a collection of handwritten samples on

a white-board acquired with the E-Beam system.

The data is stored in XML format which, in

addition to the transcription of the text, also

contains information and demographic data about

writers. The database includes 221 authors

contributing a total of more than 1,700 forms with

13,049 lines of labelled texts and 86,272

occurrences of words from a dictionary of 11,059

words. In addition to the recognition of on-line

writing, the database was also used for on-line

writer identification and gender classification

from handwriting. The collected data is stored in

XML format, and available in tif-images.

 Rimes (Augustin et al., 2006): it is an off-line

database composed of emails sent by individuals

to companies or administrations. It contains

12,723 pages corresponding to 5605 emails of

1,300 volunteers. Collected pages scanned and

annotated in RIM format, database is fully used

VISAPP 2020 - 15th International Conference on Computer Vision Theory and Applications

744

for the evaluation of tasks such as document

layout analysis, handwriting recognition,

identification and verification of writers,

identification of the logo and the extraction of

information.

 NIST: The National Institute for Standardization

and Technology (NIST) developed a series of

databases (Wilkinson et al., 1992) containing

handwritten characters and numbers supporting

tasks such as field isolation, detection and

removal of boxes in forms, segmentation and

character recognition. The NIST Special Database

1, includes samples provide by 2100 authors. The

latest version of the database, the Special

Database19

, includes handwritten forms of 3600

writers with 810,000 images of individual

characters with their meta-data. This database has

been widely used in a variety of handwritten

figures and character recognition systems.

 MNIST

: An off-line database for handwritten

digits. It is a subset of a larger set available from

NIST, includes a training set of 60,000 examples

and a test set of 10,000 examples. This database

has been widely used in machine learning and in

several digit recognition systems.

 CEDAR CDROM1 (Cheriet et al, 1994): An off-

line database contains 5632 city/state handwritten

words, 4938 handwritten states, 9454 postal

codes. Which presents a total of 21,179

manuscript digits and 27,837 alphanumeric

characters. This database has been used for the

evaluation of several systems including

handwriting segmentation, recognition of cursive

digits character recognition and word

segmentation. A CD-ROM 2 contains machine-

printed Japanese character images.

 IRESOFF IRONOFF (Viard-Gaudin et al., 1999):

A handwriting on/off database, for each character

or handwritten word, both online and offline

signals are available. It consists of 4086 single

digits, 10,685 isolated lower-case letters, 10,679

isolated capital letters and 31,346 words from a

197-words lexicon (French: 28,657 and English:

2 689).

 UNIPEN

((Guyon et al., 1994): Proposes an online

database for Western handwriting, since 1993,

contains more than 5 million Western characters,

written by more than 2,200 writers.

 IFN/ENIT (Pechwitz et al., 2002): an offline

Arabic handwritten words database was

developed from the contributions of 411

https://www.nist.gov/srd/nist-special-database-19

http://yann.lecun.com/exdb/mnist/

volunteers each filling a specific form. The

database presents a total of 26,400 words (city /

city names) corresponding to 210,000 characters.

And the meta-data includes information about the

sequence of character shapes, baseline, and

author.

 CENPARMI (Alamri et al., 2008): an offline

database for Arabic handwriting recognition. It

consists of single digits, letters, strings and words,

written by 100 participants from Canada and 228

participants from Saudi Arabia. The database has

been used for the recognition of Arabic characters,

digits and word marks.

 ADAB (Abdelaziz et al., 2014) (Abed et al.,

2009): an on-line database proposed in 2009. It

contains Arabic words corresponding to the

names of Tunisian towns and villages,

manuscripts by more than 173 different authors,

totalling more than 29,922 Arabic words and

157,792 characters.

 AltecOnDB (Abed et al., 2010): An on-line

handwriting database. The collected samples are

complete sentences that include numbers and

punctuation marks. Samples handwritten by more

than 1000 writers with different origins, genders

and ages. The collected data is used in various

tasks such as annotation and verification,

recognition, and segmentation tasks.

 OHASD (Elanwar et al., 2010): An on-line

database of Arabic handwritten sentences, the

dataset includes 154 paragraphs written by 48

writers aged 24-40 years of both sexes. It contains

more than 3,825 words and more than 19,467

characters. The dataset is used in a word retrieval

system and extraction of lines of text, as well as

word segmentation and annotation.

3 OUR CONTRIBUTION

3.1 Handwriting Database

Recommendations

Our contribution consists to propose a new database

for diacritical Arabic character recognition systems,

named "CHAKEL-DB". To achieve this goal, there

are some recommendations to be respected, whether

for the data structure or for the storage structure. Such

as: expressiveness, accessibility, parsability,

serializability, well defined and well formed.

http://hwr.nici.knu.nl/unipen

CHAKEL-DB: Online Database for Handwriting Diacritic Arabic Character

745

Following our comparative study conducted on

some standard formats (Table2), the most

recommended data formats in our context are the

inkML

and UPX

which are in an XML format. The

inkML standard presents a set of tags that describing

the electronic trajectories of writing, hence the name

Ink and considering the medium level of

expressiveness we plan to complete it with the UPX

standard which resembles the hwDataset standard

(Bhaskarabhatla and Madhvanath, 2004) with the meta

information and the annotations of the presented data

and in addition, it can detail several levels of hierarchy

for the trajectory unlike the hwdataset which presents

a single level.

3.2 Data Collection Scenario

For the collection of samples, we adopted the scenario

presented by the Figure 1, which consists in the

acquisition of the data, the collection of the trajectory

points characteristics, the capture of the character

image, and finally the annotation and verification of the

data to feed our database CHAKEL-DB.

Figure 1: Data collection scenario.

For the acquisition of the character we will need of

course a tactile surface. We qualify the trajectory

acquired by three events (Figure 2): Action_down

which indicates the beginning of the writing of the

trace, Action_move which represents the movement of

the stylet and the action_up which indicates the end of

the trace.

Figure 2: Event Recovery.

We characterize each point of the trajectory by its

coordinates (X, Y), the force F of the writing and the

timestamps. The result vector goes through a

resampling step and a smoothing step. Capturing the

result character image is useful for various uses at

online and even offline diacritic Arabic character

recognition systems. The method used for framing the

character image is based on the selection of the

endpoints (Figure 3).

Figure 3: Framing the character image.

The concepts (Figure 4 and 5) highlighted in our

sample database are the writer who defines himself by

Table 2: Handwriting Data Formats overview.

Data Format

Binary

Format

Requirements

Expressivity

well-

defined

Storage Structure

Data

structure

Accessibility

Well-

formed

Parsability Serialisability

Microsoft Ink Serialized

Format (ISF)

Yes Low Low Low Low Yes Medium

InkML No Medium High High High Yes High

Standard UNIPEN Format No Medium High Medium Low Yes High

HandWriting

Dataset(hwDataset)

No High High High High Yes High

UNIPEN XML Format

(UPX)

No High High

High High Yes High

http://www.w3.org/TR/2006/WD-InkML

http://unipen.nici.ru.nl/upx/index.html

VISAPP 2020 - 15th International Conference on Computer Vision Theory and Applications

746

his gender, his age, Hand writing, level of education,

region, type of material used and level of use, type of

script as well as the level of mastery of the script.

A writer can write several characters, each

character identified by an unique identifier that

presents his class, it is composed of several traces

each one is characterized by its identifier, its duration

and the speed of writing and it comes in a set of

points.

Our database considers 21 diacritic models for

each of the 28 Arabic characters with the 10 digits,

we will have at all: 598 models.

We characterize each point of the trajectory by its

coordinates (X, Y), the force F of the writing and the

timestamps. The result vector goes through a

resampling step and a smoothing step. Capturing the

result character image is useful for various uses at

online and even offline diacritic Arabic character

recognition systems. The method used for framing the

character image is based on the selection of the

endpoints.

Figure 4: Chakel-DB concepts.

Figure 5: Character traces example.

3.3 Experimentation and Evaluation

Our experimental task for the evaluation of this

database is in three stages (Figure 6): The first step is

to implement the mobile frame that presents the

sample collection interface, their processing to

generate a set of JavaScript Object Notation (JSON)

data that will be sent to the dedicated server. The data

retrieved at the server level by a web services

framework, which serves as a parser for JSON

messages, to generate the inkML, UPX, and tif-files

that feed our database. The second step is the

establishment of an Arabic character recognition

system to test the performance of our database.

Finally, the performance test phase.

Figure 6: Architecture of the mobile sample collection

framework.

3.3.1 Examples of Samples

As example of samples from our database, I present

this list of UPX files(Figure 7) with their

complementary inkML (Figure 8) which presents the

characters figured in the tif-images of figure 9. Our

data base presents also a set of binary images (Figure

10).

Figure 7: UPX Files.

Figure 8: inKML Files.

An example that presents zoom on the manuscript

character online “Ha” (Figure 11) with an extract of

CHAKEL-DB: Online Database for Handwriting Diacritic Arabic Character

747

the character metadata as inkML file (Figure 12) and

an extract of the Ha character metadata at the UPX

file level (Figure ) which refers to the tif-image of the

character as well as the inkML complement.

Figure 9: tif images.

Figure 10: Binary images.

Figure 11: tif-image of the diacritic character "Ha".

Figure 12: The metadata of the character "Ha" at the level of the inKML file.

Figure 13: The metadata of the character "Ha" in the UPX file.

VISAPP 2020 - 15th International Conference on Computer Vision Theory and Applications

748

3.4 CHAKEL-DB Statistics and

Comparison

In this section, we list some statistics about the

collected data in our online database for diacritical

Arabic characters. We collected handwriting samples

from 60 different writers composed of men and

women from various professional cadres,

qualifications and ages. We were interested in

keeping track of the writer's gender, age and whether

they were right-handed or left-handed. CHAKEL-DB

database presents 70% of data to training set, so for

this set we have 55 writers, 2205 Characters, and

12238 traces.

Table 3 presents a comparison database with the

other databases of the literature based on the: content

level sample numbers, file type and tasks supported.

We remark that: (i) our database deals with the case

of the diacritic sign, the others do not. (ii) it complies

with the recommendations of the online databases,

and (iii) It can be used for any application field for

CRS. The only limit mentioned is the number of

samples that we expect to increase soon.

4 CONCLUSION

In this paper we have introduced CHAKEL-DB, an

on-line database for handwriting. One of the main

objectives of this project is to collect a huge database

of handwritten characters from several writers,

presented by mobile application as input in real-time.

Unlike currently available Arabic databases, our

database: (i) includes high coverage for all possible

diacritical marks of the Arabic language. (ii) Data are

collected from many writers with different ages and

work experiences. (iii) The collected samples are

available in two levels: characters and traces (iv) and

writing involves a multi-colored, multi-sized brush.

CHAKEL-DB contains 3150 patterns of diacritic

characters and 17095 traces. For each element of the

database, a corresponding meta-data file is available

in an inKml format and an upx format. The database

is available on-line for various research applications

such as on-line recognition of Arabic handwritten

diacritical characters, on-line

identification/verification of writers, on-line

handwriting segmentation, gender classification and

on-line color identification. CHAKEL-DB will be

great help and value for the research community, and

can be downloaded by requesting a copy from LR-

SITI Lab.

REFERENCES

Abdelaziz, I., Abdou, S., 2014. AltecOnDB: A Large-

Vocabulary Arabic Online Handwriting Recognition

Database. arXiv preprint.

Abed, H. A., Margner, V., Kherallah, M., Alimi, A. M.,

2009. Online Arabic Handwriting Recognition

Competition. ICDAR.

Abed, H. A., Kherallah, M., Margner, V., Alimi, A., 2010.

On-line Arabic handwriting recognition competition,

ADAB database and participating systems.

International Journal on Document Analysis and

Recognition (IJDAR), Springer, 14, 15-23.

Alamri, H., Sadri, J., Suen, C., Nobile, N., 2008. A novel

comprehensive database for arabic offline handwriting

recognition. In Proceedings of the 11th Intl. Conference

on Frontiers in Handwriting Recognition.

Augustin, E., Brodin, J. M., Carre, M., Geoffrois, E.,

Grosicki, E., Preteux, F., 2006. RIMES evaluation

campaign for handwritten mail processing. In

Proceedings of the Workshop on Frontiers in

Handwriting Recognition.

Bhaskarabhatla, A. S., Madhvanath, S., 2004. An XML

Representation for Annotated Handwriting Datasets for

Table 3: CHAKEL-DB database characteristics.

ADAB

(Abed et al., 2009)

OHASD

(Elanwar et al., 2010)

AltecOnDB

(Abdelaziz et al., 2014)

CHAKEL-DB

Level of

Contents

Word (names of cities),

Character,

Trace, No diacritics

Sentence,

Word, Character

No diacritics

Sentence,

Word, Character

No diacritics

Character, Trace,

Diacritic signs

Number of

samples

173 writers,

29,922 words,

157.792 characters

1000 writers,

152,680 words,

6444,530 characters

1000 writers,

152,680 words,

6444,530 characters

68 writers,

3150 Characters,

17095 traces

File Type

Upx, inkml,

tif-image

Txt, isf,

jpg-image

dhw,

relational tables

(SQL DBMS)

Upx, InkML,

tif image,

binary images

Supported

tasks

Segmentation,

SRC

writer identification online.

SRC

Segmentation,

Annotation

Annotation and

verification, recognition

and segmentation

All fields of

application a SRC

CHAKEL-DB: Online Database for Handwriting Diacritic Arabic Character

749

Online Handwriting Recognition. In Proceedings 4th

Int'l Conf. on Language Resources and Evaluation.

Biadsy, F., Saabni, R., EL-Sana, J., 2011. Segmentation-

free online Arabic handwriting recognition. In J.

Pattern Recognit. Artif. Intell,25, 1009-1033.

Cheriet, M., Thibault, R., Sabourin, R, 1994. A multi-

resolution based approach for handwriting

segmentation in gray-scale images. In Proceedings of

IEEE International Conference on Image Processing.

Elanwar, R., Rashwan, A., Samia, A., 2010. OHASD: The

First On-Line Arabic Sentence Database Handwritten

on Tablet PC World Academy of Science. Engineering

and Technology International Journal of Computer,

Electrical, Automation, Control and Information

Engineering.

Guyon, I., Schomaker, L., Plamondon, R., Liberman, M.,

Janet, S., 1994. UNIPEN project of on-line data

exchange and recognition benchmarks. In Proc. of the

12th International Conference on Pattern Recognition.

Jäger, S., Manke, S., Reichert, J., Waibel, A., 2001. Online

handwriting recognition the npen Recognizer. In

IJDAR, 3, 169-180.

Liwicki, M., Bunke, H., 2005. Iam-ondb - an online english

sentence database acquired from handwritten text on a

whiteboard. In Proceedings of the Eighth International

Conference on Document Analysis and Recognition.

Marti, U. V., Bunke, H., 1999. A full english sentence

database for off-line handwriting recognition. In

Proceedings of the Fifth International Conference on

Document Analysis and Recognition.

Nakkach, H., Hichri, S., Haboubi, S., Amiri, H., 2016. A

Segmentation Free Approach to Strokes Extraction

from Online Isolated Arabic Handwritten Character. In

International Conference on Advanced Technologies

for Signal Image Processing ATSIP.

Pechwitz, M., Maddouri, S., Maergner, V., Ellouze, N.,

Amiri, H., 2002. Ifn/enit database of handwritten

arabic words. In Proceeding Of CIFED.

Plamondon, R., Srihari, S. N., 2000. On-line and off-line

handwriting recognition: A comprehensive survey. In

IEEE Transactions on Pattern Analysis and Machine

Intelligence, 22.

Steinherz, T., Rivlin, E., Intrator, N., 1999. Offline cursive

script word recognition-a survey. In Int l. J. Document

Analysis and Recognition, 2, 90-110.

Tagougui, N., Kherallah, M., Alimi, A. M., 2013. Online

Arabic handwriting recognition: a survey. In

International Journal on Document Analysis and

Recognition (IJDAR), 16, 209-226.

Viard-Gaudin, C., Lallican, P. M., Knerr, S., Binter, P.,

1999. The IRESTE On/Off (IRONOFF) Dual

Handwriting Database. ICDAR.

Vinciarelli, A., 2002. Survey on off-line Cursive Word. In

Recognition Pattern Recognition, 35(7), 1433-1446.

Wilkinson, R., Geist, J., Janet, S., Grother, P., Burges, C.,

Creecy, C., Hammond, B., Hull, J., Larsen, N., Vogl,

Y., Wilson, C., 1992. The First Census Optical

Character Recognition Systems Conference. In the U.S.

Bureau of Census and the National Institute of

Standards and Technology.

VISAPP 2020 - 15th International Conference on Computer Vision Theory and Applications

750