v1.0.1

Fixed:

• Version in ocrd-tool.json now updated

v1.0.0

Changed:

• Port the processors to use the OCR-D/core v3 API, #44
• Spawn background processes for segmentation and recognition, #44
• Refactor tests and test under various conditions (w/o METS caching, page parallel processing), #44

v0.4.1

Fixed:

• recognize: improve baseline/polgon robustness

Changed:

• segment/recognize: adapt and bump to 🔥 Kraken v5
• adapt to Python importlib instead of pkg_resources
• update tests/CI

v0.4.0

Fixed:

• binarize: OCR-D conformity (PAGE output, AlternativeImage input/output)
• docstrings

Added:

• recognize: param overwrite_text
• segment: param overwrite_segments
• segment: param level-of-operation (now supports table and region, too)

Changed:

• binarize: 🔥 renamed level-of-operation=block to region
• segment: existing segmentation will be masked away (unless overwrite_segments)

v0.3.1

Fixed:

• recognize: only apply one_channel_mode (whether to use binarized input)
  if the model has only one input channel
• recognize: project text results to region level in order
• recognize: iterate line results via proper word splitting
• recognize: add proper line and word confidences
• recognize: avoid invalid polygons on single-glyph words
• recognize: workaround for better quality box cuts
• recognize: workaround for empty/failed line records
• segment: update blla.model URL (master→main)

Changed:

• recognize: pass lines in baseline format if any baselines are annotated
• recognize: ensure baselines are valid and consistent with line polygon
• recognize/segment: fall back to CPU if CUDA not available

v0.3.0

Fixed:

• recognize: clip bounding boxes to canvas
• Updated README to reflect current capabilities
• Docker build working and with proper tagging

Added:

• resources in ocrd-tool.json for the OCR-D resource manager

v0.2.0

Added:

• text recognition with ocrd-kraken-recognize, #33
• rewrite of ocrd-kraken-segment, #33