LegatorOCR 0.2.0-pre.2

Pre-release for testing — not a stable version. Clean-room Rust post-processor for scanned books: split → deskew → crop → clean/binarize → PDF with an OCR text layer. GUI in 8 languages (ro, en, fr, it, es, pt, de, hu) and command line. MIT licence, runs locally, no accounts.

Downloads

Pick the archive for your system; every archive has a .sha256 next to it.

System	File	Notes
Linux x86-64 / ARM64	*-linux-gnu.tar.gz	glibc 2.35 or newer (Ubuntu 22.04, Debian 12 and later); OpenGL 2+, X11 or Wayland
Windows x86-64	*-windows-msvc.zip	no Visual C++ Redistributable needed; OpenGL 2+
macOS Apple-silicon / Intel	*-apple-darwin.zip	unsigned: right-click → Open on first launch

OCR needs tesseract (with the language packs you want) installed separately.

New since 0.2.0-pre.1

• Co-author: Alexandru Dumbravă (HUB DIGITAL) — the new features follow his digitisation workflow.
• Full-text search: legatorocr --search "text" FOLDER… finds phrases in PDF text layers, .txt, hOCR/HTML, .docx and inside .zip archives, ignoring case and diacritics (--search-regex, --search-exact).
• --docx: editable Word export of the OCR text. With --lang ron, ş/ţ are corrected to ș/ț.
• Profiles flux-1-optim, flux-2-draft, flux-3-volum (optimal / draft / high volume).
• Archives as input: .zip, .cbz, .tar, .tgz built in; .7z, .rar, .cab, .iso, … through an installed 7-Zip / unrar / bsdtar / unar — images and PDFs inside are processed or searched directly.
• OCR languages: two-letter codes (--lang ro+en), --ocr-languages status of the eight interface languages, install hints.
• GUI: Undo / Redo (Ctrl+Z, Ctrl+Y) for erase zones.
• Calibration on 750+ real book and magazine pages: photographs and line drawings are no longer rotated by a false skew; heavy text columns are no longer mistaken for pictures; low-dpi scans are read correctly.

Known limits

• PDF/A-2b is not certified (no veraPDF validator was available).
• JBIG2 Huffman coding and some JPEG 2000 code-block styles (bypass, termall, POC/PPM/PPT) are decoded only through an installed pdftoppm, mutool or Ghostscript.
• Only the empty user password is tried on encrypted PDFs.
• Archives: zip and tar/tgz were tested; 7z and rar were NOT tested (no extractor was available to the author). Search, Word export and the new profiles were verified on Linux x86-64 only (OCR with the English model; the Romanian model was not available to the author). The Word file opens in LibreOffice and python-docx, not tested in Microsoft Word. The text search of PDFs not written by LegatorOCR was not tested on ABBYY/PDF24 files. Verified by hand on Linux x86-64 only; the Windows, macOS and ARM64 builds come from CI and were checked for format and dependencies, not run by the author. The GUI was smoke-tested, not exercised in depth.

Documentation and presentation: https://legatorocr.chiuta.ro · Source and comparison with other tools: README.md, COMPARISON.md.

LegatorOCR 0.2.0-pre.1

Pre-release for testing — not a stable version. Clean-room Rust post-processor for scanned books: split → deskew → crop → clean/binarize → PDF with an OCR text layer. GUI in 8 languages (ro, en, fr, it, es, pt, de, hu) and command line. MIT licence, runs locally, no accounts.

Downloads

Pick the archive for your system; every archive has a .sha256 next to it.

System	File	Notes
Linux x86-64 / ARM64	*-linux-gnu.tar.gz	glibc 2.35 or newer (Ubuntu 22.04, Debian 12 and later); OpenGL 2+, X11 or Wayland
Windows x86-64	*-windows-msvc.zip	no Visual C++ Redistributable needed; OpenGL 2+
macOS Apple-silicon / Intel	*-apple-darwin.zip	unsigned: right-click → Open on first launch

OCR needs tesseract (with the language packs you want) installed separately.

New since 0.1.0

• Native JPEG 2000 decoder (JP2 and raw codestreams) for scanned PDFs, checked against OpenJPEG.
• JBIG2: generic refinement, refined text regions, aggregate symbols, pattern dictionaries and halftone regions.
• Optional PDF/A-2b structure (--pdfa).
• Adversarial hardening (size/time caps, overflow fixes) and a CI-gated release pipeline.

Known limits

• PDF/A-2b is not certified (no veraPDF validator was available).
• JBIG2 Huffman coding and some JPEG 2000 code-block styles (bypass, termall, POC/PPM/PPT) are decoded only through an installed pdftoppm, mutool or Ghostscript.
• Only the empty user password is tried on encrypted PDFs.
• Verified by hand on Linux x86-64 only; the Windows, macOS and ARM64 builds come from CI and were checked for format and dependencies, not run by the author. The GUI was smoke-tested, not exercised in depth.

Documentation and presentation: https://legatorocr.chiuta.ro · Source and comparison with other tools: README.md, COMPARISON.md.