home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
Whiteline: delta
/
whiteline CD Series - delta.iso
/
dl_serie
/
updates
/
105
/
ocr_14b
/
ocr_e.doc
< prev
next >
Wrap
Text File
|
1995-11-25
|
52KB
|
1,190 lines
============================================================================
OCR V1.4 (May 1995)
Optical Character Recognition for the Atari ST/STe/TT and Falcon by:
Alexander Clauss
Stresemannstraße 44
64297 Darmstadt
Germany
----------------------------------------------------------------------------
OCR is FREEWARE. It may be copied and used freely. Distribution by
BBSs and PD libraries is allowed. However, OCR may NOT under ANY
circumstances be included in CD-ROM collections or on magazine cover
disks without my prior written permission.
OCR and its documentation must always remain together and complete with
all its files and documentation. Changes to the program and documentation
is forbidden.
If you like OCR and use it regularly, you should send the author a
(voluntary) donation. 30-50 DM (or the equivalent in your currency) seems
reasonable but even if you don't send me any money at least send me a
postcard or Email so I know whether any interest in OCR exists...
New versions are available directly from the author (please include disk,
self addressed envelope and an international reply coupon to cover the
return postage or you won't get a reply). Alternatively you download the
latest release version from various FTP servers, try:
ftp.uni-kl.de (/pub/atari/misc) or
ftp.isbiel.ch (atari/diverses))
If you'd like to be informed when a new version is released you can either
send me your Email address or a SAE or postcard and I'll let you know by
snailmail...
Any donations in the United States or Canada will be handled by:
==============================================================================
Jeff Wisniewski
P.O. Box 1243
Newtown, PA 18940-0871
E-Mail Address:
---------------
Internet: jeffrey.wisniewski@microserve.com (I prefer this one)
dragons-egg@genie.com
All donations send from Canada MUST be in the form of a Postal Money
Order, otherwise the banks charge a fortune to cash a check not drawn
on the US Federal Reserve.
New updates/versions to the OCR program will be uploaded by me to
various services and sites in the United States.
If there are any people that do not have access to a modem or any type
of E-Mail service, then I will also be handling any mailing of updates
via the U.S. Mail only if you (the registered user) has sent the money
to cover the postage.
OCR UK support and update service
============================================================================
OCR is Freeware. If you like the program and use it regularly you should
contribute a voluntary donation. Between £10-20 is reasonable but even a
postcard or Email with feedback is better than nothing. Users who would
like to make a voluntary contribution in the UK should:
a) Make a cheque for payable to J. Connor for any amount
b) Be sure to include your name and address.
c) Send cheque to UK Address below
Joe Connor, 65 Mill Road, Colchester, CO4 5LJ, England
Email: jconnor@cix.compulink.co.uk
The UK support scheme does NOT make any profit. A small part of your
contribution is retained to cover the stationary, media, currency and
postage costs incurred running the scheme.
Why you should contribute
-------------------------
The best reason for contributing is to support and encourage future English
versions and support for OCR. You'll also receive a Master disk containing
the latest version and access to the free update service.
UK users who have contributed to OCR can obtain the latest English Version
from Joe Connor at any time, free of charge, by sending a Stamped Addressed
Envelope enclosing an unlabelled floppy disk.
Important! I'm naturally interested to receive bug reports messages and/or
suggestions for improvements. If you want to report a bug please always
include the following details: computer type, TOS version, Auto folder
programs and Accessories.
I'm particularly interested in constructive criticism and suggestions for
future versions!
Contact
----------------------------------------------------------------------------
Alexander Clauss
Stresemannstraße 44
D-64297 Darmstadt
Germany
Email: aclauss@rbg.informatik.th-darmstadt.de
(If you don't get a reply from this Email address try the snailmail
address).
Bank details
----------------------------------------------------------------------------
Sparkasse Darmstadt
Sort code: 508 501 50
Account number: 0115032143
Please indicate your address on the transfer and state 'OCR & Version
number' as reason for transfer.
English translation
-----------------------------------------------------------------------------
Stefan Damerau
Bergholm 31
D-24857 Fahrdorf
Email: Stefan_Damerau@sl.maus.de
!!!!!!!!!!!!!!!!!!!!!!IMPORTANT: MESSAGE LIMIT: 16kB!!!!!!!!!!!!!!!!!!!!!!
Edited by Joe Connor. Email at: jconnor@cix.compulink.co.uk
============================================================================
System requirements
----------------------------------------------------------------------------
This program runs on all Atari ST/STe/TT and Falcon computers. OCR
functions at any resolution above 496x200. 500 Kb RAM storage is sufficent
but working with scanned images needs a lot of memory, the more RAM the
better as usual! Naturally you'll also need a scanner.
Brief Overview
--------------
The program can recognise text in scanned images and save the text in
ASCII, which can be further processed using your favourite text editor.
Correction arrangements can be defined to assist with difficult to
recognise characters etc.
Individual databases can be created for each font.
There's a seperate mode to handle images from drawing packages where the
texts were written using the system and/or GDOS fonts which takes much less
computer time.
============================================================================
OCR functions
----------------------------------------------------------------------------
Most menu options can also be selected using keyboard shortcuts. The
corresponding shortcuts are indicated in brackets.
Starting OCR
============
When starting OCR from the command line using filenames these files will be
automatically loaded. The following extensions are supported:
- *.IMG (Images)
- *.LIB (Character set databases)
- *.KOR (Correction setups)
More than one filename can be entered, but only one of each type will be
loaded. If there's more than one of a type the one with the latest
time/date stamp will be loaded.
MultiTOS Drag & drop
--------------------
Under MultiTOS any *.IMG, *.LIB or *.KOR files can be dragged from any
desktop window to an OCR window and they will be loaded.
Dialogs
-------
All the dialogs can optionally be displayed in windows to avoid blocking
other processes when multitasking. Single tasking users should continue
to use normal dialogs as they are slightly faster.
All buttons are selectable using keyboard shortcuts. [Return] or [Enter]
selects 'OK', [Undo] selects 'Cancel'. All other buttons can be selected
by holding down the [Alternate] key and pressing the underscored
character.
Pressing the [Insert] key inside any editable field invokes a dialog box
containing the entire Atari character set. Any desired character can be
selected using the mouse. This provides an easy method of accessing
characters not normally avaialalble from the keyboard.
Popup menus can now be navigated using keyboard commands. The cursor
up/down keys may be used to highlight the desired option which can be
selected using the [Return] key. Press [Undo] or [Esc] to exit the
popup without selection.
To move any non-windowed dialog click and drag (anywhere on the dialog
background) using the mouse. Hold down the [Shift] key at the same time
and the dialog turns invisible (apart from its border) which is handy if
you need to read the contents of the screen underneath the dialog.
Windows
-------
The following keyboard entries are available:
[Control]+U Close top window
[Control]+* Toggle between full screen and original window size
[Control]+W Cycle through the open windows
Cursor keys; the contents of the topmost window is scrolled a line at a
time in the corresponding direction.
[Shift] + Cursor keys; the contents of the topmost window is scrolled a
screen at a time in the corresponding direction.
When positioning the mouse above the upper window, this window can be
scrolled by holding down the RIGHT Mouse button and moving the mouse.
Holding down either [Shift] key increases the scrolling speed.
Image windows
-------------
The scanned image is displayed in an image window. Because scanned images
are often very large it's possible to display the image at a reduced size,
50%, 25%, 12.5% etc. Refer to 'Image size' for more details.
An icon bar has been added to the image window. The icons a provide
convienent alternative to the image manipulation menu options. Here's
the details:
1. Block mode
2. Pen
3. Eraser
4. Filter image/plock
5. Bold image/block
6. Rotate 90 degrees (anti clockwise)
7. Rotate 90 degrees (clockwise)
8. Flip vertical
9. Flip horizontal
10. View original size
11. View image at 50% (half full size)
12. ... 25% (quarter full size)
13. ... 12.5% (eigth full size)
14. ... 6.25% (sixteenth full size)
The icon bar can be toggled on and off in the 'General' dialog from the
'Options' menu if you need the screen space or simply don't like them!
Text windows
------------
After recognition text is output to this window.
Marking block; using a click and drag action the mouse can be used to mark
a block consisting of complete lines inside the text window which are
displayed in inverse video. The start/end line of the block can be adjusted
by holding down either [Shift] key and clicking to mark the new block
start/end line.
Cut block [Control]+X; the marked block is moved to the clipboard and
removed from the text window.
Copy block [Control]+C; the marked block is copied to the clipboard for use
by another application which supports the Atari Clipboard.
Paste block [Control]+V; pastes an IMG format file into the Image window.
Simple text editor functions are now included. This makes it easy to
correct errors directly in OCR. The character database isn't changed if
you edit the recognised text.
Editor functions:
Cursor keys: Cursor movement
[Control] <- and ->: Cursor jumps word by word
[Shift] <- and ->: Cursor to start/end of line
[Shift] up/down: Scroll up/down one page
[Delete]: Delete character at cursor position
[Backspace]: Delete character to left of cursor
[Insert]: Toggle between insert and overwrite mode.
Underscore cursor: Insert mode
Block cursor: Overwrite mode
Note!
If a block is selected, the next character you type replaces this
block. Imagine a marked block as a gaint cursor!
============================================================================
The OCR drop down menus
----------------------------------------------------------------------------
Desk
----------------------------------------------------------------------------
Important information about OCR and the author and programmer.
File drop down menu
----------------------------------------------------------------------------
Load image [Control]+O
----------------------
Loads an image in monochrome GEM .IMG format. If an image is already loaded
a dialog appears offering the following options:
Save; before the new image is loaded, the old image is saved.
Load; the new image is loaded. Any changes made to the old image are lost.
Cancel; the command is cancelled.
Save image [Shift][Control]+S
-----------------------------
Saves the image in monochrome GEM .IMG format.
Scan picture [Control]+R
------------------------
Selecting this to scan some text. In order to do this a GDPS-Scanner
driver must be installed. If more than one driver is available select the
desired driver. Remember OCR currently only handles monochrome images. Set
the scanner (and the driver) to sensible settings!
If you work with MultiTOS with memory protection switched on, the MP-Flags
in the header of the GDPS driver should be set to GLOBAL, because
OCR and the GDPS driver need to have access to each others memory. OCR
allows its memory to be accessed by the driver if it's installed as an
desktop accessory, you don't need to change anything!
Save ASCII [Alternate]+S
------------------------
Saves the recognised text in ASCII format. If an existing filename is
selected the text in the original file will be overwritten or the text
is appended to the selected file.
Delete ASCII
------------
The recognised text is deleted. If the text has not yet been saved, a
dialog appears offering the following options:
Save; the text is saved before it's deleted.
Delete; the text is deleted.
Cancel; the command is cancelled.
Load font [Alternate]+R
-----------------------
A font database is loaded. If the font database is already loaded in memory
and it has not yet been saved, a dialog appears offering the following
options:
Save; the current font database is saved before the new one is loaded.
Load; the font database is loaded.
Cancel; the command is cancelled.
Save font [Alternate]+W
-----------------------
The font database is saved.
Delete font
-----------
The font database is deleted. If the font database has not been saved a
dialog appears offering the following options:
Save; the current font database is saved then removed from memory.
Cancel; the command is cancelled.
If you wish to create a new database the old one must first be removed from
memory.
Quit program [Control]+Q
------------------------
Exit the program using this option. If any data has not yet been saved, a
dialog appears offering the following options:
Save; a last chance to save your data.
Cancel; a last chance to remain in OCR.
End; quit OCR and return to the desktop.
Process drop down menu
----------------------------------------------------------------------------
Mark all [Control]+A
--------------------
If the Image Window is topped; the entire image is marked as a block
If the Text Window is topped; the entire text is marked as a block
Remove block marks [Shift][Control]+A
-------------------------------------
Removes the block markers from the topped window (either Image or Text).
Cut block [Control]+X
---------------------
This option is only available if the GEM Clipboard is available (although
the block will be removed anyway!) The clipboard must also be active in the
'General' dialog from the 'Options' drop down menu.
The marked block in either the Text or Image windows is moved to the
clipboard and removed from the text window.
Copy block [Control]+C
----------------------
This option is only available if the GEM Clipboard is available. The
clipboard must also be active in the 'General' dialog from the 'Options'
drop down menu.
The marked block in either the Image or Text windows is copied to the
Clipboard for use by another application which supports the Atari
Clipboard protocol.
Atari Clipboard support
-----------------------
OCR supports the Atari Clipboard which normally uses a folder on your boot
partition/disk, for example C:\CLIPBRD. Any Atari application which
supports the Clipboard can easily exchange data. The Cut and Copy options
export data to the clipboard and the Paste option imports Image data into
OCR.
Here's how it might be used in practice; After scanning in an article from
a magazine using an application eg 'SCAN.PRG' we call the Copy option in
SCAN.PRG to copy the scanned image to the Clipboard in IMG format.
Next we switch to OCR, either directly in a multitasking environment or by
leaving SCAN.PRG, and running OCR. In OCR we top the Image window and
select 'Paste' to copy the image into OCR. After the text recognition is
complete the text window is topped and using the 'Copy' option the ASCII
format text is exported to the Clipboard. Now we switch to a word processor
to edit/spell check the text before printing out the final result.
The clear advantage of the Clipboard is that no filenames were needed to
get from scanned image to the printed text, everything took place inside
the Clipboard folder automatically, a powerful tool!
Paste block [Control]+V
-------------------------------
This option is only available if the Atari Clipboard is available. The
clipboard must also be active in the 'General' dialog from the 'Options'
drop down menu.
This option is only available if the 'Image window' is topped. If the
Clipboard contains a valid IMG file it will be pasted into the 'Image
window'.
Delete Block (Delete)
---------------------
(No Menu item for this)
The 'Delete' key removes a marked block. It won't be copied to the
clipboard.
Image size [Control]+0 to 4
---------------------------
Using this menu option the image can be displayed at various sizes (this
doesn't affect the 'Recognise text' function. Under MultiTOS or on the
Falcon the image size is also adjustable through a submenu otherwise a
Dialog box appears:
100% [Control]+0 1:1 Original size
50% [Control]+1 1:2 Width and height of the image are both halved
25% [Control]+2 1:4 One quarter full size
12.5% [Control]+3 1:8 One eighth full size
6.25% [Control]+4 1:16 One sixteenth full size
Align picture (Alternate M,N,V,H)
---------------------------------
Use this option to align text which has been scanned at 90 degrees (a very
useful option for hand scanner users) MTOS or Falcon users can use the
submenu.
Rotate +90° (Alt M) : Rotates the picture 90 degrees anti clockwise
Rotate -90° (Alt N) : Rotates the picture 90 degrees clockwise
Mirror horiz. (Alt H): If the scanner is moved in the 'wrong' direction the
picture is mirrored in memory.
Mirror vert. (Alt V): See 'Mirror horiz.'
Tools
-----
Block [Alternate]+B
Marking block; using a click and drag action the mouse can be used
to mark a block inside an Image window. If the block was not
perfectly defined fist time round it can easily be repositioned by
dragging one of the eight grab handles.
Deleting blocks (Cut block) [Control]+X; use this option to remove
unwanted lines or other distractions which cannot be recognized as
text.
Multi-column text can be processed by individually marking each
column for recognition.
Pen [Alternate]+P
The image can now be edited! The line thickness is set in the
'Tools' dialog in the 'Options' menu.
Eraser [Alternate]+X
Use to erase part of the image. The size of the eraser is set in
the 'Tools' dialog in the 'Options' menu.
Filter [Control]+F
The complete image or marked block is filtered. Any dots smaller
than the threshold value set in the 'Tools' dialog in the 'Options'
menu are deleted.
Bold image [Control]+B
The complete image or marked block is thickened making all lines
and dots wider. This can help salvage scans with poor contrast.
The direction to thicken the image is set in the 'Tools' dialog in
the 'Options' menu..
Open text Window [Control]+T
----------------------------
Opens the text Window where the recognised text is displayed.
Start recognition [F1]
----------------------
Begins the Optical Character Recognition. If a block is marked only the
contents of the block is analysed by the text recognition function. If no
block is marked the entire image is analysed. It's important to remove
extraneous lines (including underlines in some cases) and other details
before starting so the individual characters can be recognised.
Important! Please ensure the correct font database is loaded. If no font
database is available, use the 'Delete font' option, if available, to
remove any font currently in memory. Select the 'Font' option from the
'Options; drop down menu and enter any relevant scanner settings into the
dialog and set the width setting for the 'Space' character (see 'Font').
The program searches for the first line. A line must have at least one
empty (white) pixel above and below the characters which must at least 5
pixels high (I'm assuming nobody expects to be able to recognise letters
smaller than 5 pixels high).
Using these rules OCR can ignore underlined words automatically although in
some case the descenders will cause problems. If this proves to be the case
you'll have to remove the underlining manually using either the 'Block
delete' or a suitable Art package.
After OCR indentifies a line, the individual characters are isolated and
compared with the characters in the font database. The character from the
database which best fits the character being analysed is inserted in the
text Window. If there isn't a sufficiently similar character in the
database to the character being analysed, within the preset degree of
saftey, one of two alternatives occurs, dependent one the setting of 'Learn
mode' option in the 'Text recognition' dialog from the 'Options' drop down
menu:
a) If 'Learn mode' is active the mystery character(s) is displayed in a
dialog box and marked in the image Window. Enter the appropriate ASCII
character into the dialog box. Up to five ASCII characters can be entered
because during scanning several characters can be merged together and in
some cases OCR cannot separate them. Joined characters (Ligatures) can also
be handled using this method. The following options are also available:
Include in database (default active); the character is included in the
database for future use. Bear in mind that a bad scan may throw up
character definitions you'll never need again, so use your judgement to
decide whether to save the character for future use.
Import into text (default active); decide whether the ASCII characters
should be inserted into the text.
Exit via 'OK' button; text recognition continues.
Exit via 'Cancel' aborts text recognition.
b) If 'Learn mode' is not active; a '?' is placed in the text.
Both the method used to compare characters and the degree of agreement
needed to consider a character as 'recognised' can be adjusted (Refer to
'Options').
Resume recognition [F2]
-----------------------
By pressing the [Esc] during recognition it's possible to break off
recognition to adjust settings and resume recognition from the last
position using the new settings, without starting from the beginning
again.
Resume recognition is only possible if the image remains unchanged so be
careful not to re-size, cut or delete the marked block, use the rotate
or mirror options or load another image. You can, however, change your
view of the image using the 'Image size' option if desired.
Break recognition [Esc]
-----------------------
Pressing the [Esc] key aborts text recognition at the end of the current
line, so it may take some time before recognition stops.
Options Drop down menu
----------------------------------------------------------------------------
Help [Help]
-----------
ST-Guide format on-line context sensitive help is available. You have
to tell OCR the location for ST-Guide using the 'Help Accessory' and
'Index file' in the 'General' dialog in the 'Options' menu.
Select ST-GUIDE.ACC and the index file OCR.REF (from the HELP folder of
OCR or in your GUIDES folder).
If you want to use OCR.HYP outside OCR, you'll have to merge OCR.HYP
and OCR.REF to the hypertext tree of ST-Guide. (refer to the ST-Guide
manual). After merging the REF file you no longer need to set the
'Index file'.
Once set-up help is available inside any dialog by pressing the [Help]
key (A 1st-Guide hypertext is available on request).
Text recognition [Alternate]+O
------------------------------
A dialog appears in which the following settings influence text
recognition:
Learn mode; if a character is not recognised using the preset levels of
certainty, this option decides the next action which can either:
a) If Learn mode is active; a further dialog box appears to teach OCR about
the unrecognised character. Refer to 'Recognise text' for more details.
b) If 'Learn mode' is not active; a '?' is placed in the text.
Agreement; this option sets the degree of agreement neccessary before a
character is considered recognised.
Sound bell on query; selecting this option sounds a ping when the end of
the text or an unrecognised character is reached. If the font database is
fairly complete, you'll hardly ever hear the ping. Deselect this option or
use the Control panel to disable the ping if it annoys you.
Display training dialog:
If this option is selected (checkbox crossed) every time an
unrecognisable character is encountered the 'Indentify character'
training dialog appears ready for you to identify the character and add
it to the database, if the character comprised more than one character
(eg ligatures) these can also be entered.
Recognition can be speeded up by de-selecting this option (checkbox
empty) in which case OCR highlights the unrecognised character and you
can enter the ASCII character directly using the keyboard. If you need to
enter more than a single character (eg ligatures) press the [spacebar] to
display the 'Indentify character' training dialog.
Insert in database/text:
Sets the default action when a character isn't recognised. Using a
large database OCR is only likely to inquire when scanning errors are
encountered. Normally you wouldn't want to include these characters in the
font database and using this option you can switch it off permanently.
Important! When you're in the process of creating a new database remember
to switch this option on again!
Character comparison; OCR attempts to recognise characters by placing the
electronic equivalent of a 'stencil' over each character. Three levels of
analysis are available:
Exact; This takes a longer processing time than medium or fast due to
different priorities for the differences when checking the
characters but yields the highest level of accuracy.
Medium; This option does not assign different priorities to differences.
It's quicker but may incorrectly recognised when scanned at a low
scanning resolution. At 300dpi this shouldn't be a problem and is
worth a try before resorting to 'Exact'.
Fast; Similar ro 'Medium' but a bit faster and probably a slightly
higher level of wrongly recognised characters.
Image quality; if the image originates from a scanner you should always
select 'Scanner'. Small differences between the characters, which occur
during scanning, are filtered using this setting.
Using the Scanner setting ALL differences between characters are
considered. As the size of the characters approaches the lower limit (5
pixels high is the absolute minimum) the filters remove so much of the
original character that recognition becomes unreliable and very time
consuming.
If the image file originates from a drawing or DTP application and includes
GDOS or system fonts these characters are likely to be precise so select
the 'Precise' setting.
Image slope
This option determines whether OCR will accept angled images. You can
usually leave this option on unless the text consists of short lines with
lots of preceding spaces. For example, envelopes with sender in the upper
left hand corner and the address in the lower right hand corner.
Line separation:
Usually (if 'precise' is set) OCR tries to find complete white lines
within the scan to seperate text lines. This won't work if the
distance between lines is very small or upper/lower parts of letters
(like g j q h or the capitals) are close together. Setting the line
separation to 'medium' or 'tolerant' may fix these problems.
'Medium' allows for a few of those cases but may also help if the
scan contains vertical black lines. This setting doesn't totally
disable line separation.
If you still have problems with line separation, try 'tolerant'.
Note!
This setting can create different recognition errors. Sometimes
(although rarely) this can also occur using the 'medium' setting.
Basically a narrow line spacing has the potential to cause problems.
Font [Alternate]+F
------------------
This option displays the options for the currently loaded database:
Minimum character size:
Width, Height: These set the minimum size a character can occupy. Any
part of the image smaller than the values set are
ignored. Small dots caused during scanning are simply
ignored.
Measure...: A crosshair shaped cursor appears, using a click and drag
action, over the text image, draw a box to define the
width and/or height of the minimum character size then
release the mouse button.
On exiting the dialog OCR checks these settings against all the
characters in the database to make sure no characters smaller than the
entered settings (otherwise they would not be checked). If any
character(s) smaller than the selected settings are found the settings
are changed to the size of the smallest character. An alert box informs
you if this has taken place.
Minimum space width:
Width: Sets the minimum width occupied by a space character.
Variable: If the font is monospaced (i.e. each space character is the
same width) deselect this option, otherwise leave it active.
Measure: A crosshair shaped cursor appears, using a click and drag action,
over the text image, draw a box to define the width of the space
between words and release the mouse button. If you change the
image viewing size the space width is automatically resized i.e
The size of the box relates to the image size)
Important! This setting determines the MINIMUM width for spaces.
Comments; use these fields to enter any desired text describing the
database, for example: font origin (magazine/book/typed) scanner settings
(dpi/contrast/brightness). This text is stored with a .LIB extension along
with the font database for easy future reference.
Edit font [Alternate]+E
-----------------------
Select this option to display the 'Edit font' dialog. Using the options in
this dialog you can undertake changes to fine tune the font database.
Font overview; using this option you can scroll through the characters. A
character can be selected by clicking on it and it appears in the 'Selected
character' area.
Selected character; this area is used to assign or change the ASCII
character (or string of up to 5 ASCII characters) associated with the
selected character. Using the 'Delete' button you can delete the selected
character from the font database.
Search (for character); this option is used to move to the desired area of
the ASCII table. 'Search' begins searching from the beginning, 'Again'
continues searching from the current position. If the search is successful,
the character found will be displayed in the 'Selected character' area.
Sort; the font database is sorted.
Statistics; an overview of the font database is displayed. From this
display it's easy to see at a glance which characters have not yet been
defined (or duplicated) along with other useful information. Only the main
ASCII characters are displayed.
Correction [Alternate]+C
------------------------
Selection this option to display the 'Correction' dialog.
Text recognition is not an exact science and is rarely perfect. There are
characters which even the human eye has difficulty recognising except by
context, eg '1'and 'l'. OCR is intelligent enough to consider the
surrounding characters and will for example return '1' if the character
being analysed is surrounded by numerals. Some characters are recognised as
two or more individual signs, <"> for example is recognised as two <'>
characters.
Using the options in the dialog these problems can be minimised.
Substitutions; enter a character string combination which should be
replaced by another. You may like to define the previous example which
converts <'><'> to <">.
Corrections; selecting this option enables the auto correction of very
similar characters such as 1/l, l/I or 0/O.
Important! Bear in mind auto-correction is not foolproof and can in
individual cases return the wrong result!
Text format:
Original format; a line of text in the image becomes a line of ASCII text,
preceding space characters can be ignored.
New format; the text format of the image is ignored and a completely new
format is imposed according to the 'Line length' setting. This is
particularly useful when converting narrow columns of text which often
contain a liberal sprinkling of hypenated words.
Remove hyphens; this option attempts to reconstruct whole words. If you're
converting German text which follows the 'k-k⇨ck' convention the additional
option to convert these characters should be activated. For all non-German
texts this option should be turned OFF.
Important! Generating a new format is not a good idea for program listings,
source code, tables etc!)
Number of empty lines; set the number of lines in the ASCII text between
seperate OCR runs.
Unknown character; defines the character used to replace unrecognised
characters if 'Learn mode'.
General [Alternate]+A
---------------------
Dialogs:
Dialog boxes can be set to appear centred on the screen or under the
mouse pointer.
The 'In windows' option (unsurprisingly) puts the dialogs into windows.
In a multitasking environment the 'In windows' option allows other
processes to continue whilst waiting for user input whereas non-windowed
(modal) dialogs freeze all other processes.
If you're NOT multitasking the normal (modal) dialogs are slightly
faster.
Paths & Files; click on any box and use the file selector to set the
corresponding path. On starting OCR these paths are automatically set so
that loading/saving takes place to the desired directory.
Text Window; if this option is active the text window is opened
automatically after starting OCR.
Clipboard; sets up which actions the clipboard will handle. This works with
both the 'Image' and 'Text windows. This is handy to 'Cut/Copy' recognised
text to the clipboard and NOT 'Cut/Copy' parts of the graphic (maybe images
apart from the scanned text) over the text in the SCRAP.TXT (There's only
ever one SCRAP.* in the clipboard). AND: Not copying images to the
clipboard saves time.
Files; click on these boxes and using the file selector select any files
you would like automatically loaded by OCR. You could use this option to
load a default set of font database and/or set of correction rules.
Tools
-----
Pen:
Sets the pen size in pixels.
Eraser:
Sets the eraser value in pixels.
Filter:
All pixel clusters smaller or equal in diameter to the set value
are deleted from the image/block.
Thicken:
This function can work in four directions up, down, left and
right. Horizontal thickening is generally more useful. This
feature is useful to repair white areas in letters which otherwise
cause errors.
ASCII format
------------
To make editing the scanned text as easy as possible control over the
line feed (LF) and carriage return (CR) characters has been
implemented.
-Atari/DOS: CR+LF.
-Unix: LF
-Macintosh: CR
The choice is yours! Some text editors only can reformat text when the
end of a paragraph uses a line feed (LF) only. This only happens when
OCR reformats the text.
System font
-----------
Any available GDOS screen font, including Speedo and NVDI 3, can be used
at any supported point size in the text window.
Save
----
Saves OCR's settings. The following settings are saved:
* All settings in the 'Recognition', 'Image size' and 'General' dialogs.
* Window layout and size for both the image and text windows.
============================================================================
Hints & tips
----------------------------------------------------------------------------
When creating a font database always be sure to record the scanner
settings. When you want to reuse the font database remember to set the
scanner to the settings used when the database was created. If you don't do
this the characters can easily appear thicker or thinner and can make the
font database virtually useless.
Ideally keep a seperate database for each different font and point size you
scan regularly. This keeps the databases small and considerably speeds up
text recognition. It's probably not practical to stick rigidly to this rule
as you may want to convert text containing both normal and italicised
versions of the same font within a body of text...
Fill out the 'Comments' field including the scanner settings, font details
and source conscientiously.
Make the best possible scan by adjusting the scanner settings. Adjust the
brightness and contrast so that characters neither 'grow' together or start
to break up ans show gaps.
If you can't get a perfect scan it's preferable to accept characters
which grow together rather than single characters which are starting to
break up, try using a higher contrast.
The lower the scanner resolution the smaller the
memory requirements for the database and the faster text recognition takes
place. However, if you scan at too low a resolution characters will start
to grow together. There's no substitute for experimentation to find the
best combination of settings for each task.
If your scanner only supports low resolutions (e.g. 200dpi or less) it
takes more effort and time to get good results. Whilst building the
database we strongly suggest you set the agreement to 90%. You'll find some
characters (eg 'e', 'c', 'b' and 'h') may well be entered into the database
more than once. After all important letters are in the database you can
lower the agreement to around 85%-75% and get acceptable results without
OCR asking too often.
By adjusting the 'Agreement' setting it's possible to get good results even
at low scanner resolutions. If characters are repeatedly being incorrectly
recognised increase the degree of agreement, although OCR is likely to
display it's manual recognition dialog more often.
Switch off Clipboard support if you don't have a harddisk (Refer to the
'General' dialog in the 'Options' drop down menu).
============================================================================
Technical
----------------------------------------------------------------------------
OCR supports:
- The expanded WM_ARROWED and WM_BOTTOMED calls in Winx 2.2.
- The VSCR Cookie, this means that dialogs will always be displayed in the
visible portion of a virtual screen.
- On the Falcon (>=TOS 4.02) and under MultiTOS the 3D effects are supported
in dialog boxes.
- Sub menus (for Image size, Align image)
- MultiTOS Drag&Drop (ARGS)
- OCR directly supports the Freedom file selector protocol (Freedom is a
non modal file selector for TOS).
- Long filename support, where available, is implemented, this also works
with MagiCMac.
RAM usage: Most RAM is used scanning the images. This RAM is needed twice,
once in standard format (the ocr accesses the data from this part) and
again in the format relevant to the display device used. Here's a guide:
A DIN A4 page (8*11 inches), scanned at 400 dpi uses around 1.7Mb.
OCR needs 2 x 1.7Mb = 3.4 Mb.
============================================================================
Known bugs and limitations
----------------------------------------------------------------------------
In rare cases OCR cannot separate two characters where OCR believes they
are only one character. For example where characters consist of several
unconnected parts and all letters followed by a punctuation mark [i, j, ä
etc.] or [ñ, å etc]. OCR does not try to seperate these letters into
several parts, however OCR does attempt to seperate several overlapping
letters although the degree of success is variable!
At high scanner resolutions the size of the database and computer time
needed becomes unmanageable, especially on a standard ST.
The size of the characters is currently limited to a mazimum of 96x96
pixels, which is enough to handle characters at 400 dpi with most scanners.
(2-3 character ligatures still fit into 96*96 pixels).
An error in operating systems <=TOS 2.06 can cause OCR to display pixel
garbage instead of a large picture. An error in the 'vr_trnfm()' function
causes this. There's a patch by Laurenz Pruessner, S. and W. Behne that
fixes this (Released in the German ST-Magazin 9/91). If you have a
TT/Falcon or use an NVDI/VDI driver with a graphics card you don't need
this fix.
Important! From OCR V1.2a this bug won't appear even without one of the
above fixes -a little trick makes this possible :-)
============================================================================
The future
----------------------------------------------------------------------------
The following enhancements are planned (suggestions are always gratefully
received):
* Analysis of unrecognised characters after the complete image (block) has
been processed. OCR could attempt to automatically complete partial
databases asking the occasional question before reaching a final
conclusion.
* Batch mode: the ability to automatically handle different texts one using
a batch file (requires the first ToDo topic)
* Allows pixel level editing of the characters in the data base.
* Handle several font databases simultaneously. This would make it much
easier to handle multi font/point size text layouts.
============================================================================
Other
----------------------------------------------------------------------------
Recommended AUTO folder programs:
COMPOSE (CKBD Deluxe 1.3) by Pascal Fellerich: This tool enables you to
choose different ways for creating special characters when editing
text. (even those which normally cannot be accessed from the keyboard).
For example you can get the '»' character by pressing the alternate key
followed by '>' twice. A mighty mouse speeder is included too. The
speed up can be set in wide ranges and several settings are available.
Winx by Martin Osieka. With the help of this program up to 40 windows can
be open simultaneously. Many other window enhancements are built-in.
All these programs are available on BBSs and FTP Servers or direct from the
authors.
============================================================================
Credits
=======
Many thanks to:
Jean-Jacques Ardoino: French RSC
Joe Connor: English RSC, English DOCs editor, UK Support
Stefan Damerau: English DOCs, all the hard work before Joe edits it!
Henk van der Griendt: Dutch RSC, LEESMIJ
Antonio Lillo: Spanish RSC, LEEME
Many thanks to all the people who made suggestions, sent bug reports or a
donation And (last but not least) to all Beta testers.
============================================================================
History
----------------------------------------------------------------------------
Version Date Changes
----------------------------------------------------------------------------
V1.4 15.05.1995 - OCR supports VA_START message (AV protocol)
- Bug within the 1/l correction fixed
- Text editor implemented
- Improved line separation
- Long filename support
- Basic image editing tools implemented
(Pen, eraser, filter and thicken)
- Optional icon bar added to image window.
- ASCII save options:
CR (Mac)
LF (Unix)
CR+LF (Atari, DOS)
- Any available GDOS font can be used for text
window display including vector fonts available
with SpeedoGDOS or NVDI 3
- IMG saving optimised (faster)
- Text window output optimised
- Various small enhancements
- OK, I admit it, many minor bugs fixed!
V1.3c 23.10.1994 - In resolutions using 8x8 pixel characters the 3D
effects (AES 3.40 or later) in dialogs could no
longer be used (needed to much space).
- Alert boxes are now automatically assigned
shortcuts.
- Two typos removed from the Dutch RSC file.
- Bugfix to MultiTOS Drag & drop protocol handling
- Problems with Let'em Fly and OCRs Windowed
dialogs resolved. OCR now handling the
extended-ob_type root objects in the dialogs and
Let'em Fly is instructed to keep out!
- Bugfix: Saving the parameter sometimes didn't
work if shel_find() could not find the file.
Possibly caused using a an alternative file
selector which changed the actual path and also
under MultiTOS and Geneva.
V1.3b 20.09.1994 - Bugfix: Error with line seperation removed (This
could lead to bombs under certain circumstances).
V1.3 10.08.1994 - Keyboard shortcuts in dialogs.
- Bugfix: Bug in Searching for keyboard shortcuts
in the menu fixed.
- OCR now reacts faster to events during recognition.
- Windowed dialogs ([Undo] same action as 'Cancel')
- ST-Guide AND 1st-Guide format on-line Help provided.
- Bugfix: Problems with 'I/l Correction' resolved
- GDPS scanner driver support
- Two buttons moved from 'General' to a better
position in 'Recognition'.
- Characters can be indentified without displaying
the 'Indentify character' dialog which is much
faster as [Return] after each character is no
longer needed
- Minimum character measurements can be set which
results in better separation of characters and
pixels in bad scans)
- Filter can be switched ON/OFF
- Checkboxes and radio buttons can be selected by
clicking the text beside them.
- Problems with some IMG's (those which contain more
scanlines than defined in the header) fixed
- New file format for *.LIB and *.KOR files.
Naturally the old format is still supported
V1.25b 30.05.1994 - Bugfix: Bug in character separation fixed
- Bugfix: Bug in 'Reformat' (Correction dialog)
fixed
V1.25a 20.05.1994 - Bugfix: German RSC-file: Text of an Alert button
shortened :-(
10 characters per button is the limit and I hadn't
noticed this because MultiDial and Let'em fly
both support more.
- Bugfix: Error handling 'Save ASCII' fixed
- Bugfix: MTOS Problems when scrolling (apparently
MTOS sends MU_BUTTON-events too)...
- Info line in the graphics window. The image and
block (if marked) sizes are displayed. During
recognition progress is reported. When running
under in a multitasking environment the busy bee
is no longer displayed and control is enabled.
- When running under in a multitasking environment
the some menu items can now be selected during
recognition.
V1.25 05.05.1994 - 'Resume recognition' now possible
- Better recognition of <'> or <'> characters
- 'k-k⇨ck' correction (hyphen removal option) can be
switched off (important for all non-German texts)
- Filenames: Missing extensions automatically added
- Filenames (IMG, LIB, KOR) can be loaded automatically
by passing them as parameters from a command line
- Line seperation bug removed (caused some crashes)
- Newer, faster algorithm for comparison using the
'medium/fast' setting in the 'Recognition' dialog
- AP_TERM (MTOS) support
- Recognition speed dramatically increased
- Image alignment (mirror/rotate) built-in (ready for
direct OCR scanner support)
- Moveable window contents using right mouse button
- MultiTOS Drag&Drop support (ARGS)
- English DOCs included.
(Many thanks to Stefan Damerau and Joe Connor)
V1.2a 29.03.1994 - Character seperation bug removed (only caused
problems with 'thin' GDOS fonts; This bug was built
in with V1.2 first time :-/
- VDI font size calculation bug removed
V1.2 08.03.1994 - Moveable dialogs can be move transparently (by
holding down the [Shift] key)
- Faster recognition
- Better character separation
- Wildcard for unknown letters now user defined
- IMG load error fixed
- Oblique scans can now be recognised
- Several minor new features added
- First release with English manual/Resource and French
resource
V1.1a 06.02.1994 - Unneccessary 'redraws' removed
- Text block deleting error fixed
- Error in character comparison removed
V1.1 04.02.1994 - Clipboard support added
- Redraw error in scrolling fixed
V1.0a 01.02.1994 - Redraw error in the graphics window fixed
- OCR doesn't try to install a submenu under Mag!X 2.0
V1.0 30.01.1994 - First release
----------------------------------------------------------------------------
EOF