3dpcp/.svn/pristine/42/42bac048a870ae1be3532211aa99ccf24bb0c7a7.svn-base

66 lines
2.9 KiB
Text
Raw Normal View History

2012-09-16 12:33:11 +00:00
TODO LIST
Please send any ideas.
Next release (0.4x)
- quality test script for groups of samples (to ensure improvement in recognition)
single chars or single chars in words, clean and difficult sets,
formatted texts
false negative + false positive
start with numbers? option numbers only?
output: find fname.{jpg,png} + fname.txt + compare output gocr against txt
options: [testpath] [testchars]
bin/gocr_chk.sh testbase/free/{clean,glued,dusty}/{numbers,text}/
bin/gocr_chk.sh jocr/examples tmp09
- fix problem with cutting melted chars (using vector frames)
- better detection of agglutinated serifs (Gutenberg scans 086.png+171.png)
- vectorize recognition (big step!, relation to other OSS?)
(find min distance to ideal vector patterns, start with <>())
- frame_nn is marking only the borders like frame_vector and removed later
- handle broken and glueed chars by the database algorithm (-m 256 -m 130)
- improve get_line2(), implement distance_to_point and distance_to_line
- dot-matrix printouts (examples/matrix.jpg) (german: Nadeldrucker)
- examples/inverse.pcx + examples/rotate45.pcx by nearest-box-to-line alg.
or mean nearest box (or its 4 edges) directions,
rotate only boxes (by creating new greater boxes and tread as new image)
- proof replacement of getTextLine by getXMLline via pipe(?) or stdout
is pipe available on all platforms?
- docu about ispell using via XML (what needed, test in gocr.tcl)
- replace rest of UNDEFINED in unicode.c by its correct strings
- add probability for box->m1..m4 (to reduce errors caused by bad line-scan)
call line detection function second time to improve unsure line data
Next release (0.5x)
- reduce pixel data by vectorization (big change, faster)
- writing images through pipes (like reading)
- using dictionary (optional) for replacing not recognized chars
- Karsten.Hilbert@gmx.net: use ORChie WordBox-format
see http://http.cs.berkeley.edu/~fateman/kathey/ocrchie.html
aspell instead of ispell
Near future: (planned version)
- rewrite install-routine
- perspective distortion (for cameras)
- genetic algorithms engine (already in development, 0.8?). It includes
feature extraction and classification
- support for other languages (may affect context_correction(), etc) (0.6?)
- support for diagramation. Can be done using the Unicode+new stuff. I (bbg)
have some ideas.
Far future:
- gimp plugin
- color support
- Braille detection (usefull for blinds?)
see: American Journal of Physics Vol. 70, No. 7, p 684-688 (2002)
or use special foils
- read image in smaller parts, to reduce memory usage.
- frames should be recognized
- better distance function (comparision of characters)
- detection of orientation (i.a. 90,180,270deg rotation)
- picture extraction
- math formula detection, font type detection
- handwritten texts (blockletters)
--- uff, really a lot of work ---
- Feel free and add your suggestions and wishes,
or tell me, what is the most important point for you.