v0.13.3

Changed:

• recognize: on glyph level, fall back to RIL.SYMBOL if ChoiceIterator is empty

v0.13.2

Fixed:

• updated requirements

v0.13.1

Fixed:

• deps-ubuntu/Docker: adapt to resmgr location mechanism, link to PPA models
• recognize: 🐛 skip detected segments if polygon cannot be made valid

Changed:

• deskew: add line-level operation for script detection
• recognize: query more choices for textequiv_level=glyph if available
• recognize: 🔥 reset Tesseract API when applying model/param settings per segment
• recognize: 👀 allow configuring Tesseract parameters per segment via XPath queries
• recognize: 👀 allow selecting recognition model per segment via XPath queries
• recognize: 👀 allow selecting recognition model automatically via confidence

v0.13.0

Changed:

• segment*/recognize: annotate clipped,binarized AlternativeImage on page level
• binarize: add page level, make default

v0.12.0

Changed:

• resource lookup in a function to avoid module-level instantiation, #172
• skip recognition of elements if they have pc:TextEquiv and overwrite_text is false-y, #170

Added:

• New parameter oem to explicitly set the engine backend to use, #168, #170

v0.11.0

Changed:

• Models are resolved via OCR-D/core resource manager default location ($XDG_DATA_HOME) or $TESSDATA_PREFIX, #166

v0.10.1

Fixed:

• segment*/recognize: reduce minimal region height to sane value
• segment*/recognize: also disable text recognition if model is empty
• segment-{region,line,word}: apply only single-level segmentation again
• segment*/recognize: skip empty non-text blocks and all-reject words

Changed:

• segment*/recognize: add option shrink_polygons, default to false
• segment*/recognize: add Tesseract version to meta-data
• recognize: add option tesseract_parameters to expose all variables

v0.10.0

Fixed:

• when padding images, add the offset to coords of new segments
• when segmenting regions, skip empty output coords more robustly
• deskew/segment/recognize: skip empty input images more robustly
• crop: fix pageId of new derived image
• recognize: fix missing RIL for terminal GetUTF8Text()
• recognize: fix Confidence() vs MeanTextConf()

Changed:

• recognize: add all-in-one segmentation with flexible entry point
• recognize: re-parameterize to segmentation_level+textequiv_level
• recognize: 🔥 rename overwrite_words to overwrite_segments
• segment*: delegate to recognize
• recognize: also annotate orientation and skew when segmenting regions
• fontshape: new processor for TextStyle detection via pre-LSTM models
• crop: also use existing text regions, if any
• deskew: delegate to core for reflection and rotation
• deskew: always get new image and set feature deskewed (even for 0°)

v0.9.5

Fixed:

• logging according to OCR-D/core#599 (again)

v0.9.4

Fixed:

• recognize: be robust to different input image modes, Pillow#4925
• logging according to OCR-D/core#599