• improve table merge algorithm (added check on table layout) MultiPageTableExtractor.
• refactoring table merge MultiPageTableExtractor.
• improve header footer analysis HeaderFooterDetector.
• added header footer analysis support in Tabby.
• added header footer analysis info (parameter need_header_footer_analysis) in documentation (readthedocs).
• update to python3.10.
• update to ubuntu22.04.
• added Contributing Information

• Added simple multilingual textual layer correctness classification based on letter percentage calculation (textual_layer_classifier=letter).
• Added a new parameter textual_layer_classifier = [simple, ml (default), letter].
• Remove parameter fast_textual_layer_detection. Now it is a textual_layer_classifier=simple.
• Fix bug with table_type=table_wo_external_bounds (fixed cv2.BoundingRect).
• Some refactoring TableRecognition.
• Added parameter table_type and TableRecognition info into documentation.

• Upgrade PyPDF2 to pypdf>4 and fix bugs in attachments extraction from PDF files.
• Added each_page_textual_layer_detection parameter for textual layer detection on each page of PDF documents (for PdfAutoReader).
• Added ENABLE_CANCELLATION env variable for enabling/disabling parsing cancellation after client disconnection (enabled by default).
• Fixed location coordinates of attached images extracted by PdfTabbyReader.
• Added new reader PdfBrokenEncodingReader for PDF documents with textual layer but broken encoding (pdf_with_text_layer=bad_encoding).

• Improve merging multi-page tables in PdfTabbyReader.
• Stop parsing after client disconnection (for API usage, see issue 488).

Release note: v2.3.1

• Fix bug with bold lines in DocxReader (see issue 479).
• Upgraded requirements.txt (beautifulsoup4 to 4.12.3 version).
• Added support for external grobid (added support parameter Authorization).
• Added GOST (Russian government standard) frame recognition in PdfTabbyReader (need_gost_frame_analysis parameter).
• Update documentation (added GOST frame recognition).
• Added multi-page table handling to PdfTabbyReader.

• Dedoc telegram chat created.
• Added patterns parameter for configuring default structure type.
• Added notebooks with Dedoc usage (see issue 484).
• Fix bug OutOfMemoryError: Java heap space in PdfTabbyReader (see issue 489).
• Fix bug with numeration in DocxReader (see issue 494).
• Added GOST (Russian government standard) frame recognition in PdfImageReader and PdfTxtlayerReader (need_gost_frame_analysis parameter).

• Fix bugs with start, end of BBoxAnnotation in PdfTabbyReader.
• Improve columns classification and orientation detection for PDF and images (is_one_column_document and document_orientation parameters).
• Upgrade docker: docker-compose is no longer supported, use docker compose instead.
• Fix bug of tables parsing in DocxReader (see issue).
• Added simple textual layer detection in PdfAutoReader (fast_textual_layer_detection parameter).
• Improve paragraph extraction from PDF documents and images.
• Retrain a classifier for diplomas (document_type="diploma") on a new dataset.

• Upgrade dependencies: numpy<2.0 and dedoc-utils==0.3.7.

• Added internal functions and classes to support integration of Dedoc into langchain
• Upgrade some dependencies, in particular, xgboost>=1.6.0, pandas, pdfminer.six

• Show page division and page numbers in the HTML output representation (API usage, return_format="html").
• Make imports from dedoc library faster.
• Added tutorial how to add a new language to dedoc (not finished entirely).
• Added additional page_id metadata for multi-page nodes (structure_type="tree" in API, TreeConstructor in the library).
• Updated OCR and orientation/columns classification benchmarks.
• Minor edits of README.md.
• Fixed empty cells handling in CSVReader.
• Fixed bounding boxes extraction for text in tables for PdfTabbyReader.