v0.19.1

Fixed:

• Correct version in ocrd-tool.json

v0.19.0

Fixed:

• segment*/recognize: delegate process instead of overwrite for docstring
• segment*/recognize: more robust polygon handling

Changed:

• more and more concrete tests
• 🔥 require Shapely v2
• Update tesseract to 5.4.1, #214

v0.18.0

Changed:

• tesseract and tesserocr included as submodules, installable via make instal-tesser{act,ocr}, #197
• Updated docker setup accordingly, #197

v0.17.0

Fixed:

• segment/recognize: fix shrink_polygons
• segment/recognize: fix reinit scope (for xpath_model and auto_model)
• CI: test multiple Python versions independent of ocrd/core image
• CI: speed up build for EOL Python 3.6
• CI: chmod o+w tessdata directory of PPA/OS Tesseract
• deps-ubuntu: allow installation of PPA Tesseract to fail (for newer OS)

Changed:

• adapted to Shapely v2
• *: inherit from recognize (but override logger)
• segment*: delegate constructor instead of wrapping instance
• requires ocrd==2.48

v0.16.0

Changed:

• require newer OCR-D/core to include OCR-D/core#934, #188
• no more need to set TESSDATA_PREFIX
• improved and up-to-date README

v0.15.0

Added:

• binarize: dpi numerical parameter to specify pixel density, #186
• binarize: tiseg boolean parameter to specify whether to call tessapi.AnalyseLayout for text-image separation, #186

Changed:

• regonize: improved polygon handling, #186
• resources: proper support for moduledir, companion to OCR-D/core#904, #187

v0.14.0

Changed:

• list all resources in the ocrd-tool.json, #184, OCR-D/core#800
• custom --list-resources handler, #176

v0.13.6

Fixed:

• segment/recognize: no find_tables when already looking for cells

Changed:

• segment/recognize: add param find_staves (for pageseg_apply_music_mask)
• segment/recognize: 🔥 set find_staves=false by default

v0.13.5

Fixed:

• recognize: prevent invalid empty Unicode glyph choices

v0.13.4

Fixed:

• recognize: only reset API when xpath_model or auto_model is active
• recognize: for glyph level output, reduce choice confidence threshold
• recognize: for glyph level output, skip choices with same text
• recognize: avoid projecting empty text results from lower levels

Changed:

• recognize: allow setting init-time (model-related) parameters