1st Canadian Shareware Disc/1st_Canadian_Shareware_Disc_1991.ISO/graphics/ocr_vec/ocr_vec.doc

Text File | 1990-05-10 | 13KB | 337 lines

        OCR_VEC         Copyright (C) by Ron Mignery, 1990



Introduction:

        OCR_VEC is a trainable optical character recognition system for
        the pc.  It enables the computer to scan a graphics screen
        image in order to recognize and to translate whatever text it
        finds.  The graphics screen image would normally come from a
        hardware scanner device, though its actual source is not
        important to the software.  Test images could be generated with
        a paint program, for example.

        This package is intended only as a demonstration of a novel
        approach to optical character recognition.  The classical
        approach follows the "eye" model.  The entire character image
        is compared against a library of stored images to find a
        match.  OCR_VEC's approach follows a "Braille" model.  The
        computer feels the shape of the character and matches the
        vectors generated against a library of stored vectors.  Vectors
        are perhaps more easily scaled and abstracted than images and thus
        this system is somewhat more tolerant than the classical approach
        to variation in letter shapes and sizes.

        The software can be trained on any font or fonts.  When the
        computer scans an unrecognized letter, it displays the letter
        and prompts for user input.  The user may then identify the
        letter and add its vectors to the lookup library.  In
        subsequent encounters, the computer will now recognize that
        letter.


The OCR_VEC package:

        The OCR_VEC packages contains the following files:

        OCR_VEC.DOC     This file
        OCR_VEC.EXE     The OCR executive program
        CGA_CAP.EXE     A TSR utility to capture CGA screens
        EGA_CAP.EXE     A TSR utility to capture EGA screens
        HGA_CAP.EXE     A TSR utility to capture HERC screens
        VGA_CAP.EXE     A TSR utility to capture VGA screens
        OCR_SHP.SHP     A sample letter shape library file
        OCR_LET.LET     A sample letter description file
        OCR_SCR.BIN     A sample graphics screen capture file
        *.BGI           Turbo C graphics drivers


Operation:


Optical Character Recogniton:


1.      Enter OCR_VEC at the command line.  The program will then prompt:

            Enter name of shape file (default=OCR_SHP.SHP)


2.      Press the ENTER key to accept the default name or type
        in the filename of your choice and then press ENTER.

        The known letter shapes are read from this file.  If the file
        does not exist, the file will be created but no known shapes
        will be loaded.  All new letter shapes encountered during the
        scan will be written to this file.


        If the shape file is empty, the program next prompts:

                Lookup file not found!

            Enter granularity (default=8.0)

3.      Press the ENTER key to accept the default value or type
        in the value of your choice and then press ENTER.

        The granularity value determines how precisely the computer
        feels the shape of each letter.  If this value is too large,
        the shape library becomes too large and performance is
        compromised.  If this value is too low, ambiguities in letter
        recognition occur.  The program may be unable to distinguish Q
        from O for example.  A value of 8.0 means that any feature of a
        letter must be larger than 1/8 the size of the entire letter to
        be considered significant.  Empirically for normal fonts this
        value is sufficient.  Use a larger number if ambiguities in
        character recognition occur.


        The program next prompts:

            Enter name of letter file (default=OCR_LET.LET)

4.      Press the ENTER key to accept the default name or type
        in the filename of your choice and then press ENTER.

        This file contains information about each letter in the ASCII
        set and about user specified doublets and character fragments
        described below in the Adding a character section.


        The program next prompts:

            Enter screen image file (default=OCR_SCR.BIN)

5.      Press the ENTER key to accept the default name or type
        in the filename of your choice and then press ENTER.

        This file must exist and be in the format required by OCR_VEC.
        A sample file is included with the system.  To generate your
        own files from your own graphics, use the capture programs
        described below.


        The program then loads the requested graphic screen and
        presents the prompt:

            Scan entire screen (Y/N)?

6.      If you want to scan only a portion of the screen, press N and
        follow the instructions to mark off the portion of the screen
        to be used.


        The program next begins scanning for characters.  The progress
        of the scan is indicated by screen inversion.  If the upper
        left pixel is off, the program assumes a black background and
        scans for non-black characters. If it is set, the program scans
        for black characters.  Scanning stops when a blank line is
        encountered following non-blank lines.  The program then
        highlights a character and prompts as follows:


        If the character is found in the shape library:

                Is it _ ? (Y/N/P/A/D/ESC)

        Else:

                Not found.  Add to list (Y/N/A/D/ESC)?

7.      For the first prompt:

                Type Y if the character highlighted is the
                character scanned.  Note that only a portion
                of the scanned character may be highlighted and
                that the highlighted portion may be identical in
                shape and location to the suggested character.
                The / in % for example. Type Y if / is suggested.

                Type N otherwise.  Do this only if the highlighted
                part is of a different shape or location than the
                suggested character.

                Type P to purge the indicated letter from the
                shape library.

        For the second prompt:

                Type Y to add its shape to the shape library.
                Follow the directions for Adding a Character
                outlined in the next section.

                Type N otherwise.

        For both prompts:

                Type A to toggle auto mode.  When auto mode is on, the
                first prompt is bypassed.  Use this mode when all
                characters in a font have been loaded into the shape
                library.

                Type D if the line being scanned has no letters with
                descenders (letters that extend below the baseline).
                This rare occurrence can confuse the system.  An asterisk
                will appear in the prompt box when this mode is active.

                Type ESC to quit.


8.      Repeat step 7 until the entire text line has been scanned.

        The program will then present the text of the line just scanned
        and the following prompt:

                <Text of line just scanned>

            Press any key to continue...(A/O/D/ESC)


9.      Type A to toggle auto mode as described above.

        Type O to append the line of text to the file OCR_VEC.OUT.

        Type D to rescan the line in no descenders mode as described
        above.

        Type ESC to quit.

        Type any other key to continue.


10.     Continue this procedure until the entire screen has been
        scanned.


Adding a Character:

        When you type Y in response to the Add to list
        (Y/N/A/D/ESC)?  prompt described above, the program then
        prompts:

                Enter letter that was scanned.

        You should then enter the correct case letter that is
        highlighted.  Since OCR_VEC identifies a character as any group
        of non-background color pixels that is completely surrounded by
        background color pixels, the progam will sometimes identify
        parts of characters or multiple characters as a single
        character.

        In particular, the dot above i and j will be treated as a
        single character and will be identified before the body of the
        letter since the program scans downward.  The dot and all other
        letter fragments should be treated as special cases and
        identified to the program through control characters.  That is,
        if the program highlights the dot above i or j, then ctrl-a
        should be entered as the letter scanned.  You will then be
        able to enter two letters to correspond to the highlighted
        character.  Similarly, the tops of the two-part characters ?
        and ! should be entered as ctrl-b and ctrl-c respectively.  The
        dots above : and ; may or may not be identified as the same as
        the dots above i and j depending on the font scanned.  If they
        are not the same, then ctrl-d should be used.  These
        assignements are abitrary and are only suggestions.  The point
        is not to assign multiple values to a single shape.  Later,
        when the program scans the second part of the character, it
        will recognize the first part above it and combine the
        information to generate a single character output.  Parts are
        combined only when one part completely overlaps another part.
        Characters that partially overlap will not be combined; thus,
        even a kerned font can potentially be read.

        Some legitimate fonts have no whitespace between certain letter
        combinations like fl or YZ.  OCR_VEC will highlight two
        characters in such a case.  Again, these combinations can be
        handled with control characters.  Enter an unused control
        character to correspond to the doublet. You may then enter two
        letters to represent the scanned character.  A little awkward
        but it works.  If a font has no whitespace between lines,
        however, you are out of luck.



Graphics Screen Capture:

        OCR_VEC requires screen image files in a special format.  Files
        in this format are generated by the ?GA_CAP.EXE programs
        included in the package.  These programs are TSRs (terminate
        and stay resident) that are invoked after loading by the
        simultaneous press of the Alt, Ctrl and S keys.

To capture a screen into the file OCR_SCR.BIN:

1.      To capture CGA graphics screens, enter CGA_CAP at the Dos prompt.

        To capture EGA graphics screens, enter EGA_CAP at the Dos prompt.

        To capture VGA graphics screens, enter VGA_CAP at the Dos prompt.

        To capture HERCULES graphics screens, enter HGA_CAP at the Dos
        prompt.

        Load only one TSR at a time.  These TSRs are quick and dirty
        little affairs and must not have other TSRs loaded after them.
        Use of Mark and Release utilities is recommended for their
        management.


2.      Run whatever program you use to display the graphic screen
        you want OCR_VEC to scan.


3.      Simultaneously press the Alt, Ctrl and S keys.


4.      The file OCR_SCR.BIN is created in the current directory.



Copyrights:

        OCR_VEC Copyright (C) Ron Mignery 1990.

        Created using Turbo C, Copyright (C) Borland 1987, 1988.
        Graphics drivers included as permitted by Borland's
        No-Nonsense License statement.

        TSRs created using the KyCorp Memory Resident Library Version
        2.01, Copyright (C) KyCorp Information Group, Inc. 1987.


Disclaimers:

        This software and instructions are provided "as is" without
        warranty of any kind either expressed or implied including but
        not limited to fitness for a particular purpose. The entire
        risk as to the results and performance of the software is
        assumed by the user.


Distribution:

        The OCR_VEC package may be copied, distributed (free of
        charge), and used non-commercially provided that it is not
        modified in any way.

        Anyone interested in this approach to optical character
        recognition may contact the author at the following address:


                Ron Mignery

                85 Bartlett Street

                Somerville, MA 02145

                [phone]


        (or via GENIE ron mignery)

***end of OCR_VEC.DOC***