| The OCR is the Optical Character Recognition system. It
introduces the idea of comparing the scanned images with the electronic text.
Scanned images are obtained through scanners, which means that it resembles the copied
images. The scanner translates the image into a grid or a map of millions of
dots. The dot, which is referred to also as bit in computer terminology, is assigned
a value by the scanner interface. This value (for binary scanner) is either a
"0" representing a blank zone or a "1", for an inked zone.
The number of dots forming the page map depends on the scanner RESOLUTION. E.g. for
a scanner resolution of 300 dots per inch (dpi), there is about 90,000 dot per square
inch, and accordingly a full A4 paper will be formed of 8,415,000 dots. The map of
these dots (bits), and hence the names bit-map, is much like a photograph your
painting. The only way to make any changes is on the bit-level, where you can change
colors.
On the other hand, text characters are assigned Identity codes, which are commonly known
as ASC II codes. Different code sets can be assigned to characters and are referred
to as code-pages. All word processors, spread sheet, databases, and other text
processing system, basically manipulate these text codes. Modification in text files
is thus possible on the character level rather than on the bit level as in images.
OCR is thus the process of converting a bit-map of a scanned image containing text to text
codes (ASC II). At a first novice glance, the OCR process seems simple, when
compared to human reading. In computer domain, however, OCR is a sophisticated heavy
application.
Research in the OCR field, started in the late 40's, and the first OCR system (Latin)
reached the market in the 1954. Early systems were limited specially designed fonts
such as OCR-A, and OCR-B. The OCR technology has progressed in popularity and
sophistication, where by the 1970's, trainable systems appeared in the market. These
systems are designed to cover the growing variety of fonts used in typewriters and
computers.
In the early OCR technology, the recognition process is merely is between a bit-map's of
the character to a library of character shapes. This method is called PATTERN
MATCHING, and works for a specified number of fonts and sizes, although characters vary
considerably in font and size.
Later OCR technology, is based on FEATURE EXTRACTION, which identifies a character, based
on some local and topological characteristics. By starting with the bit-map of a
character, the process analyses the overall shape by extracting individual features; such
as curves, corner, and line breaks. The main advantage of using feature extraction,
is the ability to recognize several fonts and sizes, without learning each font a
priori. This technology, however, requires a relatively strong processing power, and
is implemented in most Latin OCR products.
Now, we introduce the Recent OCR technology, which called OMNI OCR engine. It can
recognize any text without the need of teaching the system the font nor take any samples
for shapes. The new OMNI engine adopts new technologies one of which is the Common
Fate Technology. This technology is based on intensive geometrical analysis of
shapes; the engine evaluates more than one possible solution - based on analysis of the
text data given-, selecting the best possible one to get the best accuracy.
Having an OMNI OCR engine did not deprive the user from the Trainability features that
were already available in previous versions of ICRA.
The nature of Arabic printed text is radically than that for Latin other languages, which
has a direct impact on the OCR process. The main technical challenge faced is the
CURSIVENESS of Arabic text, this is a fact since concatenation of isolated Arabic
characters is unacceptable, and is not a normal way of Arabic writing. Therefore,
the major problem that has to be resolved is the SEGMENTATION of words into
characters. The concern in Latin OCR is rather different, where the RECOGNITION of
characters for different fonts and sizes occupies the major focus.
Other difficulties which occur in many fonts in Arabic, is that characters OVERLAP on top
of each other, either in one solid form or overhanging on their position within the word,
which implies a larger pattern classes to be recognized relative to Latin languages.
It is important to point out that there are some Arabic characters, which are highly
ambiguous in shape
Data Entry
The major bottle - neck in many large-scale business applications nowadays, is how to get
data, embedded in documents inside the machine. This requires data to be keyed in;
and key entry was (and still is) an expensive, repetitive, time consuming process for
large amounts of text. Also recent results of ergonomic research has shown that
excessive keyboarding caused some physical problems for the data entry person. OCR
can provide a reasonable practical solution, and can be used to generate full text data
banks.
Desktop Publishing
Extracting text from input documents, which can be exported to another documents, after
being OCRed. The recognized text can be published within different environments.
Automatic Indexing
There is growing need for efficient automatic indexing of documents. Indexing
actually defines a link between the images stored, and the information those images
contain. Typically, the operator of an input station feeds documents into the
scanner and then types in a logical index. The OCR process when focussed on a
certain zone can do this task, with minimal human intervention.
Basic OCR technologyTrainable Approach extended with OMNI properties.
Even though recognition in ICRA is based on font training the OMNI approach to OCR still -
in a way - exists. This is done in two ways:
First by allowing the user to have any number of fonts active during recognition, choosing
a set of fonts can provide wide coverage for a large number of fonts. The user can
build his own library of fonts with which to recognize any new font.
Second by allowing recognition size dependent or size independent. Size independence
provides a wider coverage of fonts and sizes but at the expense of accuracy.
This compromise between OMNI and TRAINABLE technologies is an attempt to make use of the
advantages of the two approaches. The main advantage of the OMNI approach is that
almost any text can be recognized without having to train the system first on the font
used to write the text. This, however, is at the expense of decreasing accuracy for
non-standard fonts where some characters might have slight shape variances.
TRAINABLE technology solves the problem of these "unusual" fonts and a system
that uses this approach can virtually recognize any font. this, however, is at the
expense of time since each new font will first require training.
Shape Training
Recognition in ICRA is completely based on font training. Unlike the Latin character
set where each character has a unique distinct shape, the Arabic character set contains
characters that can have an identical shapes and are only differentiated by their hangers
(dots, double dots, etc…). "Fonts Training" in ICRA is done through
learning shapes rather than characters which greatly reduces the training time for a
font. If there are three characters with the same shape then only one sample for the
three characters is needed to train the system to recognize them.
Page Analysis
Page Analysis is one of the most useful features in any OCR product. The main aim of
this feature is to retain the structure of text within an image as much as possible in the
output text as well. This is very necessary in newspaper and magazines
articles. Given a page, ICRA processes the entire page and identifies zones of text
and graphics. Each zone of text is recognized as a separate entity producing a
corresponding block of text while zones of graphics are eliminated from recognition.
Adaptive Trainability
This is a feature that enhances the TRAINABILITY approach used in ICRA, even though a wide
range of common irregular shapes may appear in uncommon irregular fonts. An
irregular shape is defined to be a sequence of a number of basic character shapes.
Adaptive Trainability solves the problems of many unknown irregular shapes that may appear
in some fonts. It involves allowing the user to define any new irregular shape by
combining a number of the basic character shapes. The user can then take samples of
this new shape and the system would identify the irregular character shape. ICRA
uses an expert system based on a knowledge base of rules in its identification of
characters. The rules of any new irregular shape are dynamically acquired.
Retain Page Formatting
ICRA attempts to reproduce the original page formatting and layout as closely as possible
in the recognized document by preserving the following attributes:
Relative text column positioning
Margins
Tabs
Inter-line spacing
Indentation
Blank vertical spacing
In this chapter we will explain the use of the ICRA Gold command summery in integration
with the Universal Word 2000. The user has three menus, these menus are explained as
follows.
OCR Menu
The ICRA Gold Main menu is a part of the Universal Word 2000 menus. It appears as follows
from the Tools menu:
Where:
File/Settings: Update the ICRA
Gold settings.
Update Irregulars: Define and update irregular shapes.
Learned Font: Learn a new font to be used in recognition.
Panel: Show
the Panel window.
Mode Menu
The ICRA Mode menu option is used to login into the ICRA mode in order to start the
recognition, the learning and training processes. This menu option could be attained
through the Image Menu, as follows:
Recognize Menu
The Recognize menu option is used to start the recognition of the new font. This
menu option could be attained through the Image Menu.
SettingsICRA Gold recognizes images through two methods:
With enabling the OMNI base.
Without enabling the OMNI base.
Before recognizing any marked blocks of text on the currently loaded image, the user has
make sure that the appropriate settings are chosen. This could be done from the
Settings menu option in the File menu. In the following we will be more familiar
with the different settings options and functions.
To define or change the settings, select the Settings item from ICRA sub-menu in the Tools
menu, where the language Tab, OCR Tab and the Output Tab appear.
The language
This version of ICRA Gold recognizes the Arabic language.
To select a language, follow the next steps:
Select the required language from the Available Languages list box. The properties
of the selected language will be displayed to the right in the Properties section, these
properties are:
Direction:
Left to right
Right to left
Cursive:
Cursive
Non-cursive
Diacritic:
Language contains diacritics
Language doesn't contain diacritics.
Click the OK button.
To exit without changing the previous selected language, click Cancel.
OCR
The OCR tab is place, where the user can select the parameters that control the
recognition process of a certain image.
To define or modify the contents of the OCR tab
Select the OCR Tab from the ICRA Gold Setting dialog box.
Recognition Base
The recognition base determines the method in which the text to be recognized will be
treated during recognition. The user can either select or
Unselect the Enable OMNI Base option.
The OMNI Base
The OMNI Base is a base created from variety of samples, taken from the most general
features that a regular font could have. Those samples are taken to cover the
majority of shapes that an image could have. This option is valid only for the
Arabic language.
Click the Enable OMNI Base check box, when the user wishes to perform OMNI or partially
OMNI recognition of Arabic text.
The Font Family
This option is valid only for the Arabic language. Font family specifies the
category in which the font to be recognized falls.
Regular: Select Regular if the font is not Koufi.
Koufi : Select Koufi if the font is Koufi.
Text Nature: This option is valid only for the Arabic language. This option
specifies the source of a font by selecting the one or more of the followings:
Unknown: This option could be selected if the source of the font is unknown.
Known: This option could be selected if the source of the font is one of the
following:
Computer Output: The Computer Output is used if the font is produced from computer
printing.
Books: The Books is used, if the font comes from books.
Newspapers: The Newspapers is used, if the font source is a newspaper.
Magazines: The Magazines is used, if the font source is magazine.
Typewriter output: The Typewriter Output is used, if the font comes from a
typewriter.
Note: specifying the text nature helps in getting better accuracy in the recognition of
the selected text.
Trainable Fonts
This list box displays all the fonts for the currently selected language. If you
select a font it will be added to the recognition base while unselecting it will remove it
from the recognition base.
i.e. if the user selects one of the trainable fonts, the font will be used to complement
the OMNI Base in the recognition process.
Noise Level
This facility was provided to filter out noise in documents from the output text. If
a certain document contains a minimal level of noise this value can be adjusted to filter
out this noise. Increasing this value will filter out noise with larger dimensions
while decreasing it will filter out noise with smaller dimension. The user should handle
this option with care, however since if this value is too large part of the actual text
(e.g. dots and double dots) can be excluded from the text causing misrecognition. If
the user tries to take samples of dots in the training utility and they fail to appear in
the object window then this value is most probably too big and should be decreased.
Allow Disconnection
This option is only valid for the Cursive languages.
Select the Allow Disconnection option when you want to recognize cursive text that is
disconnected at some places due to bad quality printing.
Filter Graphics
Filter Graphics means that the system will not include any graph or drawing in the
recognition of the image.
The Output options determines the way the text from recognition will appear in the text
file.
Select the Output tab to define or modify the setting of the output text. It appears as
follows:
In the above figure, the user can edit in the output of the recognized image. The
following options are available in the above figure:
Save Output As
This is the option where the user can select the files in which the result of the
recognition will be saved. This option can be:
One File
Multiple blocks are appended one after the other in the same file.
Block Separator enabled
Any piece of text that is used as a separator between blocks of text, in a single file,
that correspond to successive blocks that have been marked or automatically
detected. This is only valid when multiple blocks are to be saved in One File.
Select block separator enabled if you wish a block separator to appear between blocks of
text.
Before Text: This option is selected if the user wants the block separator to appear
before every block of text.
After Text: This option is selected if the user wants the block separator to appear after
every block of text.
Output Format
From this option the user can select the format that the result of the recognition will be
saved in.
Text Format: The output text is a flat ASC II format.
Vertical Indentation: The vertical indentation is used to retain the vertical
indentation of the output text.
Horizontal Indentation: The horizontal indentation is used to retain the horizontal
indentation of the output text.
Un-recognized Character: From this option the user can specify the character to
appear in the output text for any un-recognized character.
Creating Trainable Fonts
The main idea behind creating trainable fonts is to enhance the recognition accuracy for
documents.
The system is designed to learn the basic character shapes of any language that has been
defined using the "Language Builder" module. A basic character shape can
be a diacritic (e.g. a Dot or a Hamza in Arabic), a special character (e.g.? or ! in
Latin) or some Arabic ligatures that has been pre-defined in the system or defined using
the Shape designer. For languages that have upper and lower case characters (e.g.
English) both upper and lower cases form of every character has to be taken.
Character Forms
Any language defined using the language builder can either be cursive or non-cursive.
All character shapes in a non-cursive language can occur only in ISOLATED form.
For a cursive language each character shape can occur in one or more of the following
positions
START
MIDDLE
FINAL
ISOLATED
Some of the shapes may be present in only one or two positions while others may occupy all
four positions.
Ligatures
Due to the complexity prevailing in the printed text of some languages a special provision
has been made in the system to handle ligatures. Ligatures are characters that override
each other in a vertical manner, which poses a challenging problem when trying to
decompose it into individual characters
Any encountered ligature can be defined using the Shape designer and hence samples of it
can be taken.
Touching Shapes
Touching shapes are consecutive character shapes that connect together when the distance
between them becomes very small. It was found that some touching shapes are very common in
some fonts of some languages.
For the shape designer, the maximum number of shapes that can constitute a ligature or
touching shape is 'Seven'.
Learning Font Window
The Learning Font Window is a window that contains all the trainable fonts available in
the system in a list. It gives the user the ability to create new trainable fonts
and add them to the font's library.
To select a font or to create a new font, select the Learning Font from the ICRA sub-menu,
in the Tools menu. As the user do that the next window appears:
The Current Font that is currently utilized in the recognition is displayed at the top of
the window. A list of trainable fonts is also displayed.
The user can add sample to the list of fonts by doing the following:
Place the cursor on the Add Sample to Font edit box.
Write the name of the desired font.
Then click OK.
Main Body Window
The Main Body Window is the window where the object, that the user have selected to take
samples from, is placed. These shapes could be either Regular or Irregular
shapes.
This window will appear, when the user double clicks on the character in the Magnifying
Glass window.
In the following figure the Main Body window appears:
In the above window the user can notice:
The constituent characters of the object will appear in Black. These are the shapes
that can be taken as samples.
The Light Gray portions of the object are discarded by the system and cannot be taken as
samples.
The Red portion of the object represents the currently selected basic shape.
The two Dark Gray portions represents regions that can be filled in black, thus joining
together the two black regions either side to form one basic valid shape.
To switch to the fill mode in order to fill these regions, click the Fill button.
Then click the required dark gray region.
To set the fill mode off, press the Arrow button .
The Cutting Options list
The cutting options list is a list that contains the cutting options available for the
current selected object (shape).
Select the option that will correctly segment the object into a number of valid character
shapes.
When you click on the button the next Selected Shapes Window shows.
To define a regular shape, the user can do the following:
Select the character(s) that should be defined by one of the following buttons:
The Cutting Options button.
The Filling button.
The Arrow button.
Click, the left mouse button, on the appropriate shape from the Panel Window.
Repeat the previous action in the rest of the shapes available in the Object window.
To define an Irregular shape the user can do the following:
Select the character(s) that should be defined by one of the following buttons:
The Cutting Options button.
The Filling button.
The Arrow button.
Define these characters from the Panel window using the right mouse button.
As you do that you will notice that the defined characters will appear respectively in the
lower part of the Main Body Window.
To add the defined shape to the Irregular shape's library, press the Add button. A
dialog box will appear asking if the user wants to define this sample.
By clicking the Yes button the Shape Designer Window, where the user can adjust the
selected shape, will appear.
To undo the last selected shape, press the undo button.
Tip: the user can cancel the whole selected shape from the lower part of the Main Body
window by pressing the Delete button
Magnifying Glass Window
The Magnifying Glass window appears as soon as you double click on any part of the
recognized text. It magnifies the same part of the text that you double clicked.
As you double click on the any object in the magnifying glass, the Main Body Window comes
into view with the selected object.
Panel Window
The Panel window is the window that contains the basic character shapes as well as the
diacritics (if any) of the currently selected language that have been defined using the
Language builder (other than Arabic). Each button contains the basic shape of a
character or a diacritic.
The different Blue color degrees on the button of the character shape, reflect the
position of the samples that have been taken for the character shape represented by that
button.
A character shape's button that appears in normal light gray color indicates that no
samples have been taken for that character shape.
A character shape's button that has in the lower half of it a blue color, indicates that
at least one sample has been taken but that they still remains a position with no sample.
A character shape's button that appears in blue color indicates that all of the character
shape's positions have been filled with samples.
Tip: Diacritics can only accept samples in ISOLATED form. Thus these character shapes
require only an ISOLATED sample for their color to switch to blue.
To select one of the basic character shapes for sample taking just click with the mouse on
its corresponding button in the panel.
Shape Designer Window
The Shape Designer window is used for designing and defining new irregular character
shapes that are unknown to the system. This increases the range of irregular
character shapes that the system can identify.
To define and design a new irregular character shape, the user must choose the sequence of
the basic character shapes from the Panel window. From the Main Body window the bitmap
associated with it will appear in the
Note: Any basic character can be used to design a new shape. Up to 7 basic characters can
be used to build a shape.
Shape Designer window
The Sample button is used to fill the Shape Designer window with the bitmap, which the
user has selected from the Main Body Window.
The Clear button is used to empty the Shape Designer from any bitmap.
The Text button is used to change the bitmap of the currently selected irregular shape in
the defined shapes first to a textual representation, which the user can specify.
The button used is to edit the bitmap representing any new shape.
The button is used to erase any part of the bitmap representing new shape. After
defining the sample, press Add Sample button to restore the designed sample in the
Irregular Shapes.
Tip: To cancel all the operation, select the Cancel button.
Diacritics
If the basic shape contains a diacritic it will be placed in the shape designer according
to some classification that depends on its position relative to the other basic
shapes. These classification are divided into 4 positions, which are:
Upper Child
In the first position, an upper child diacritic is an upper child to the second basic
shape that is, in the sequence of basic shapes, constitutes the irregular shape.
For example: The irregular shape consists of two basic shapes which are the nabra
and the triple dots.
In this case the character and his upper child are defined from the Selected Shape Window.
When the user selects the diacritic character by the right mouse click, a dialog box,
where the user can select the type the position of the diacritic, appears as follows:
In the middle position, If the diacritic is neither the first nor the last in the sequence
of shapes it can be classified as an upper child to the basic shape that comes next in the
sequence.
The first character shape above is an irregular shape that consists of a sequence of three
basic shapes, which are the nabra, the dot and the hah. The dot here is an upper
child of the hah character.
The character and his upper child are defined from the Selected Shape Window.
Lower Child
In the middle position, if the diacritic is neither the first nor the last in the sequence
of shapes it can be classified as a lower child to the basic shape that comes before it in
the sequence.
The first character shape is an irregular shape that consists of a sequence of three basic
shapes, which are the nabra, the double dots and the meem. The double dots in this
example is a lower child of the basic shape nabra.
In the last position, a lower child diacritic is a lower child to the second basic shape
that is, in the sequence of basic shapes, constitutes the irregular shape.
For example: the irregular shape consists of three basic shapes. These shapes are
the feh, the yeh and then the double dots, where the double dots is the lower child of the
yeh
.
This irregular shape is defined from the Panel window by the right mouse click, and it
appears in the Selected Shapes.
Upper Neighbor
In the first position, The upper neighbor diacritics in the first position does not belong
to the second basic shape in the sequence but it is considered an upper child to the
character that comes before it in the context of some peace of text.
The first character shape above is an irregular shape that consists of a sequence of two
basic shapes. These shapes are the hamza and the lam, where the hamza is the upper
neighbor to the lam character. Since it does not belong to it, whereas it s an upper
child to the character that comes before the lam in this particular piece of text.
In the last position, an Upper Neighbor diacritics in the last position does not belong to
the previous basic shape in the sequence, but it is considered an upper child to the
character that comes after it in the context of some piece of text.
In the above shape the last character shape is an irregular shape that consists of three
basic shapes the lam, the alef and the double dots. The double dots in this example
represents an upper neighbor diacritic to the alef character. Since it does not
belong to the alef, otherwise it is considered an upper child to the character that comes
after it in this piece of text.
Lower Neighbor
In the first position, a lower neighbor diacritic in the first position does not belong to
the second basic shape in the sequence, but it is considered a lower child to the
character that comes before it in the context of some piece of text.
In the above shape the second character shape is an irregular shape that consists of two
basic shapes the double dots, and the ein. The double dots in this example
represents a lower neighbor diacritic to the ein character. Since it does not belong
to the ein, otherwise it is considered a lower child to the character that comes before it
in this piece of text.
In the last position, a lower neighbor diacritic in the last position does not belong to
the previous basic shape in the sequence, but it is considered a lower child to the
character that comes after it in the context of some piece of text.
The above character shape is an irregular shape that consists of a sequence of two basic
shapes, which are the reh, and the double dots. The double dots in this example
represents a lower neighbor diacritic to the reh character. Since it does not belong
to the reh, it is considered a lower child to the character that comes after it in this
piece of text.
Irregular Shapes Window
The Irregular Shapes Window is the window that contains all the irregular shapes
previously defined through the Shape Designer window.
The irregular shapes window consists of three main windows, that you can observe:
Defined Shapes: It's a list showing the pictorial representations of all previously
defined irregular shapes.
Irregular Shape window: This window contains the pictorial representation of the irregular
shape that is currently selected in the Defined Shapes list.
Basic Shapes window: It shows the constituent basic character shapes of the currently
selected irregular shape in the Defined Shapes list.
The Text button is used to change the bitmap of the currently selected irregular shape in
the defined shapes first to a textual representation.
Warning: While updating the bitmap, you cannot restore the bitmap to its state before the
last changes you made.
The Clear button is used to clear the shape designer window.
The Update button is used to update the bitmap of the currently selected irregular shape
with the current bitmap in the Irregular shape Window.
The Delete button is used to delete the currently selected irregular shape in the Defined
Shapes list.
The OK button is used to close the dialog and save all the changes you made.
The Cancel button is used to close the dialog and cancel all the changes you made.
The Add Sample Dialog
The Add Sample Dialog appears only when the character shape is defined in the Manual
Learning Mode, unlike the Automatic Learning Mode, where the sample is immediately saved
and the selection is moved to the next character of the object. This dialog appears
when a letter is defined from the Panel window.
This dialog contains two windows. The left window shows the character shape taken as
New Sample, while the right window shows the previously taken samples of the same
character shape.
The user can see the currently existing samples and decide whether or not to save the new
samples.
Press the Add Sample button to add the character shape.
Press the Close button to cancel the operation.
On clicking the Add Sample button, the sample will transfer to the right window indicating
that it has been saved.
Press the Delete button to delete the current shape or the added shape.
Three small windows are installed at the right side of the window.
The first shows the Total Number of available previously taken samples of this character
shape.
The other window presents the Position of the current sample in proportion to the previous
or the next character in the specified text. The position might be a Start, Middle,
Final or Isolated (like this example).
And the third window is a miniature that shows the Shape of the current sample character
that appears in the right window.
Learning Process
To start learning a new font the user can do the following:
Open the archive and the folder where the required image exists.
Activate the image.
Select the ICRA Mode item form the Tools menu to enter in the ICRA Mode.
Define the required font from the Learning Fonts panel.
Activates the Panel Window.
Select the shapes that will be trained to the system by double clicking on it from the
Image window.
The Magnifying Glass window appears, double click on the selected shape from the
Magnifying Glass window.
The Main Body window appears with the shape in it. From the panel window click on the
character that represents the selected shape.
Repeat these steps for all the shapes that will be defined to the system.
If the shape is irregular use the irregular shapes (mentioned above) process in defining
it.
Language BuilderICRA Gold is not only an Arabic OMNI base, but also it can be trained to
recognize other languages. That's why the Language Builder exists. The Language
Builder is the utility where users can define to the system the language, basis and rules
that control the efficiency of the recognition process.
Moreover, the language builder is considered an important tool that helps the users in
enlarging the scale of recognized images, to get to the most possible accuracy in the
field of character recognition.
The Language Builder Command Summery
The Language Builder command summery, where the user can perform different manipulations,
includes:
Language Menu: The Language menu have the following:
New: Create a new language.
Load: display an existing language.
Delete: Delete an existing language
Undelete: Undelete the deleted languages
Save: Save the current active
Save As: Save the current active language
with a new name
Exit: Quit the application; Prompt to save changes
Toggle Language Menu: The Toggle Language menu helps in changing the interface language
from Latin to Arabic and vice versa. This appears in the following example:
Latin Language Arabic
Language
Define a New Language: The user can define to the system a new language, and the defined
language could be used later in the recognition of any image in ICRA.
To define a language, follow the following steps:
Select the New item from the Language menu.
Direction: The direction of the Language, that the user will read, whether it will be from
Left to Right or from Right to Left.
Cursive: A language is known to be cursive if its words are formed of connected
characters.
For example, Arabic is a cursive language, but English may not be considered cursive if we
are talking about the printed English.
Diacritic A language is known to be diacritic if some of its alphabets are distinguished
from some other alphabets (or in other words hangers) above or under them.
A language is a set of Basic Shapes, Basic Shapes Rules and Equivalence Rules. These
three components together with the samples enable ICRA to correctly recognize a text
written with this given language. The following Basic Shape window appears so that
the user can start defining the basic shapes of the defined language:
The basic shape window is the window where the user can define, add, edit and/or delete
the basic shape (characters/diacritics) that constitutes the language. The basic
shape window appears in the following figure:
As we can notice the above window appears empty, as the user did not define any shape yet.
To start defining basic shape: Press the New button, the following window will appear:
Shape Name: The user can enter the name of the shape that will be defined in the Edit
Box. This name could be any meaningful name. E.g. Aleph, Beh, The, etc…
Note: The shape Name should not exceed 15 character.
Root: From this option the user can specify that this shape is a character shape that has
a root. The root shape has valid position and invalid positions.
For example, the Aleph may come at the end of a main body or may come as a stand alone
character but it cannot come in the middle of a word. The followings are the
positions of the character shape:
Start: Select this button if the character could have a 'Start' position.
Middle: Select this button if the character could have a 'Middle' position.
Final: Select this button if the character could have a 'Final' position.
Isolated: Select this button if the character could have a 'Isolated' position.
Diacritics: From this option the user can specify that this shape is a Diacritics shape,
e.g. Dot, Double Dot, Hamza etc…. The followings are the positions of the
diacritic shape:
Up: Select this button if the character could have a 'Upper' position.
Down: Select this button if the character could have a 'Lower' position.
The Diacritics shape may come in different modes, these modes are:
Simple: This option is selected if the diacritic is in a simple form - not composed of
some other hangers, e.g. Hamza.
The simple diacritic can also come in an Isolated Form such as the Hamza, which is in some
cases a hanger and in other cases it is an isolated character. If the user is
defining a hanger that could come in both cases, then he can select the "Isolated
Form" check box, and vice versa.
Compound: This option is selected if the diacritic is in a compound form - composed from
other hangers, e.g. Double Dots which is composed from two Dots.
To define this option, drag the diacritic components from the 'Hangers' list box then drop
it in the 'Components' list box.
After defining the properties of the selected shape, the user can enter the Shape form, in
the blank box.
This shape will be represented in a bitmap form as soon as the user presses the Paste
button. The user can edit in the pasted shape, by clicking on the mouse in the
Bitmap area.
To delete a shape from the bitmaps area, press the Clear Selection button, the bitmap will
be completely cleared.
Press OK, when finishing.
The defined shape will appear in the Basic Shapes window, as follows:
To view the details of this shape, press the Details button, the Basic Shape Details will
re-appear.
The Basic Shapes Rules are those that defines alphabet characters in terms of a basic
shape and a hanger.
In this window, a Defined Rules list box showing all the available rules and a group of
check boxes that show the position, where the defined rule will be applicable, are
displayed.
To define a new rule, press the New button.
To define a new rule, select a basic shape from the 'Shapes' list.
Drag it. The following cursor appears while dragging
.
Drop it in the leftmost edit box.
Then, select the appropriate hanger from the Hangers list.
Drag it. The following cursor appears while dragging .
Drop it in the middle edit box.
Note: Beside the hangers there are arrows shown which indicates the position (up or down)
of the hanger.
Type the character that the rule represents in the rightmost edit box.
Then select the position that this character can have from the list of Positions
available.
Press OK, when finishing.
The defined shape will appear in the Basic Shapes window, as follows:
To view the details of this shape, press the Details button, the Basic Shape Details will
re-appear.
The Equivalence Rules are those rules that give ICRA a definition of the meaning of some
sequences of basic shapes.
For example, an equivalence rule, in Arabic Language, may states the following; that if a
TICK (start or middle) comes followed by another middle TICK followed by a NOON then these
three shapes represent a SEEN. This rule is very important as in the segmentation of
the main body the segment of ICRA cuts the SEEN into two TICKS and a NOON.
The equivalence rule may consist of two or three sequences. A sequence composes of a basic
shape in one or more positions, and a hanger in a certain position.
In the example stated above, we have three sequences:
TICK in Start or middle position.
TICK in Middle position.
NOON in the Final position.
To define a new Equivalence Rule, click on the New button in the Equivalence Rules window.
The Define Equivalence Rule Dialog will show up as shown below:
To compose this rule, choose the shape (Tick) from the Shapes list box.
Select one of positions (Start or Final) from the Positions list box.
From any of the list boxes while holding the CTRL key drag the selection and drop it in
the Members list box.
Repeat the above sequence for the next numbers sequences.
Press OK.
The rule now is defined and it appears in the following window:
RecognitionTo start recognizing any scanned images that exists in the archiving system as
an imported document, the user can follow the following steps:
Open the Archive, where the image exists.
Open the parent folder of the image.
Select the image from the Document window of the folder.
From the ICRA sub-menu in the Tools menu, select the Settings item to adjust the setting
of the current recognition, if desired (For more details about the Setting menu of ICRA
Gold.
Select the ICRA Mode item either from the Image menu, or from the mouse right list.
Select - part of or all - the page to be recognized by blocking it.
To block an area, place the cursor on the beginning of the text area, hold the button down
while moving the cursor to the end of the desired area, then drop the button.
Tip: To delete the selected block, select it then press the Delete button, or select the
Delete item either from the Edit menu or from the Popup menu. The selected block
will be deleted.
Select the Recognize item either from the Image menu, or from the Popup menu.
ICRA will start recognizing the selected image and the Image caption will show the
percentage of successively completed recognition.
OCR Results
The results of the recognition of an image will appear to the user as an annotation
attached to the document.
To see this result or this annotation, select the Annotations items either from the View
menu or from the Popup menu.
Select the annotation ICRAresult. Then press the Edit button to open this annotation.
If the user wants to teach the system some of the shapes that were not recognized
correctly, he has to follow the below steps:
- Select or create a font from the Learning Fonts
dialog box.
- Select the required shape from the Image window
by double clicking on it.
The Magnifying Glass window will appear with the selected shape. Double click on the
selected shape in the magnifying glass.
The Main Body window appears.
Start the training process.
|
