Introduction to OMNIOCR 2.2 - Chinese OCR for UNIX ================================================== 3/26/1997 1. Introducing myself: Hi, glad to meet you. My name is Baochun Ge, author of OMNIOCR for UNIX. We know, it's painful to input Chinese characters into computer, I bet most of us had the experience of searching around many Chinese literature archives on the internet, so often we only got a piece of the great novels, poems, etc., because the good-hearted volunteer simply cannot afford to finish the arduous input job! There are some commercial Chinese OCR software, but most of us cannot get hold to them, because they are not available from the internet. I am proud to be the author of OMNIOCR 2.1 for UNIX - the first Chinese OCR software to be archived on ifcss. 2. System requirements: SGI 4000 or 10000 running IRIX 5.x and up; SUN Spark Running SunOS 5.x and up. HP-UX workstations. with a minimum of 16 MB memory. omniocr 2.2 for UNIX does not support other platforms right now. 3. How to install and use OMNIOCR: A) Get OMNIOCR software from the Internet. ftp: ftp.ifcss.org/pub/software/unix/misc/omniocr2.2.README ftp: ftp.ifcss.org/pub/software/unix/misc/omniocr2.2.tar.gz You may try your luck at other sites too: (Beijing) info.bta.net.cn:/pub/software (partial); (Shanghai) info.sta.net.cn:/pub/software (partial); (Singapore) ftp.technet.sg:/pub/chinese (full) (Taiwan) ftp.edu.tw:/Chinese/ifcss/software (full); (Taiwan) nctuccca.edu.tw (IP: 140.111.3.21); (US) cnd.org:/pub/software (IP: 132.249.229.100); B) Unpack the software archive: csh% gunzip omniocr2.2.tar.gz csh% tar -xvf omniocr2.2.tar csh% cd omniocr2.2 (if you are working on a SGI IRIX or SGI IRIX64 system:) csh% mv omniocr.sgi omniocr (if you are working on a SUN OS systm:) csh% mv omniocr.sun omniocr (if you are working on a HP-UX systm:) csh% mv omniocr.hp omniocr If you have problem to run OMNIOCR on your machine, please contact the author. C) Get font files and unpack them: alib.24.gz (kai) blib.24.gz (fang song) clib.24.gz (song) dlib.24.gz (hei) flib.24.gz (fan) from ifcss(or it's mirror sites) in /pub/software/fonts/gb/misc directory. put the fonts in omniocr2.2/fonts directory and unpack them using gunzip. Notice that most printings use either song or kai style fonts. D) Initialize fonts: If it's first time you run omniocr, you have to initialize fonts by typing: omniocr setup this will create initialized fonts: omni24.alib omni24.blib omni24.clib omni24.dlib omni24.flib in omniocr2.2/fonts directory. Once these files are created, you can delete alib.24, blib.24, clib.24, dlib.24, flib.24 to save disk space. Warning: once directory omniocr2.2 is created, you should not change it's structure, otherwise you may have problem to run the program. E) Run omniocr: (setup alias to omniocr program, preferably in your .cshrc file) csh% alias omniocr PATH_TO_OMNIOCR2.2/omniocr (now you can run omniocr anywhere in the system) csh% omniocr -help 4. Basic features of OMNIOCR: Recognizes 5 font types: Kai, Fang Song, Song, Hei(black), Fan(traditional style); Automatic adjustment of font size, recognizes from 6 pt. and up; Recognizes both horizontal and vertical printed text; Shows both scanned and converted text on the same screen; Creates Guo Biao file and HZ file at the same time; OMNIOCR is very easy to learn to use and works extremely fast with a high percentage of recognition accuracy; Works on UNIX platforms, you don't have to have a graphic terminal in order to use OMNIOCR! 5. Input image requirements: OMNIOCR reads bi-level and grey-scale TIFF image files. The recognition accuracy is directly affected by the quality of the input image, which includes resolution of the image, skewness of the image, missing pixels(image scan too light), noisy pixels(image scan too dark), and so on. It's recommended to scan the image with 200 to 300 points per inch resolution. If the printing uses a non-standard or bad quality font, it'll also affect the recognition accuracy. As many of you may know, the recognition accuracy of any OCR software is very sensitive to the skewness of the scanned image. So it's strongly recommended that the user uses a FLAT-BED scanner, not a HAND-HELD one, in order to achieve highest recognition accuracy for your software. 6. Speed: You must worry about the speed of your software unless you own a Cray. OMNIOCR Chinese OCR engine uses a lot of advanced algorithms to speed-up the recognition process. But talk about speed without talking about the computer system is meaningless. Take SGI workstations, for example, the recognition speed could reach 15 characters/ second (system time) in a r10000 mainboard, 12 CPU system, while on a r4000, 1 CPU system, the speed could be as low as 3 characters/second(system time). 7. Copyright and stuff: The license for this trial version of OMNIOCR 2.2 for UNIX has a limited time. If you find OMNIOCR useful, you are encouraged to register with a NO fee. To register, contact the author at: e-mail: bge@ltpmail.gsfc.nasa.gov or billjr@hotmail.com By registering OMNIOCR 2.2 for UNIX, you'll get the updated information on any new releases of the software in the future. Those who are interested in licensing the source code should also contact the author directly. 8. Standard disclaimer: THE AUTHOR(S) OF THE SOFTWARE MAKE NO REPRESENTATIONS ABOUT THE SUITABILITY OF THE SOFTWARE FOR ANY PURPOSE. IT IS PROVIDED "AS IS" WITHOUT EXPRESS OR IMPLIED WARRANTY. THE AUTHOR(S) SHALL NOT BE LIABLE FOR ANY DAMAGES SUFFERED BY THE USERS OF THE SOFTWARE. END================= END =================END ====================END