3dpcp/.svn/pristine/b8/b871335920fc28139a7c592c8c222ac63470a8af.svn-base
2012-09-16 14:33:11 +02:00

57 lines
3.3 KiB
Text

This file is intended to clarify why and how we are using Unicode in gocr. It's
probably only interesting if you intend to do something similar in a project of
yours or to develop gocr.
History
0.1 initial version
---
Why to use Unicode? While in this early development stage gocr doesn't
recognize much more than the ASCII characters, we hope that someday it will
support many different languages with different character sets; that it will
recognize mathematical expressions, and so on. Even in this early stage, we are
trying to support other Latin languages --- accented characters. Since once we
aren't using ASCII characters anymore we are subject to the character set
loaded in the machine if we use the 0x80-0xFF characters, we had to solve the
problem.
Against what Andrew Tanenbaum once said, "The good thing about standards is
that there are so many to choose from", we decided to not invent a new one and
stick to one of the current; Unicode is the most famous, so we chose it.
To my dismay, Unicode's support, at this time, sucks. There are few libraries
around to deal with it, contrary to what one would expect. The libraries I
found, though very good, did not provide the kind of support we needed in gocr:
to work internally with hundreds of different characters. They were all focused
in handling external files, user interface --- i18n, in short --- something that
I'm sure is much more needed and used than what gocr needs.
That's why we wrote our own Unicode code. We implemented only what we needed,
and in a practical way to the developer --- composing characters, etc. Since no
one I know will want the output of their scanned and OCR document in Unicode or
UTF-8 format (though I hope that one format will eventually be used in every OS
and computer around, and ASCII will go to a museum, and though gocr can output
in one of these formats too), we had to output in some format more friendly;
the choices are existing character maps, TeX, SGML and HTML initially, and who
knows what else later. Once we can recognize the text and keep the formatting,
these formats will be desired even more.
How to implement it (careful: developer's view)? Fortunatedly, there is partial
support for it now. The wchar_t type defined in <stddef.h> is a standard (only
sometimes 16, sometimes 32, perhaps even 64 somewhere). Do we need the libc's
string functions? If we do, they also exist for wchar_t. Some conversion
functions were needed: ASCII -> Unicode, Unicode -> everything else.
The ASCII -> Unicode conversion (done by the compose() function) is written to
be called by the ocr engine, when it recognizes a character. You can also use
the Unicode codes #defined in unicode.h, but the compose function allows a
simpler use. It's recommended to use the symbols itself for ASCII codes (don't
need to LATIN_CAPITAL_LETTER_A, use 'A').
The Unicode -> etc conversion (done by decode()) is a bit more difficult
sometimes, since previous symbols may interact with the current one. For
example, if you're converting to TeX, two characters that are in math mode will
call two times math mode; for example, "\( \pi \) \( \iota \)", instead of
"\( \pi \iota \)". Possibly a wider conversion function, decode_text(), which
deals with the entire text at once should be provided; this function will also
create headers, etc.