Skip to content

HomeLedger v1.1.1

Choose a tag to compare

@TastingRogue TastingRogue released this 17 Sep 00:07
· 31 commits to main since this release

Patch release focused on receipt OCR accuracy for English and Spanish documents. No breaking changes — the /api/v1 surface and backups are unchanged (backup metadata version bumped to 1.1.1).

Added

  • Image pre-processing before OCR using sharp (auto-orient, grayscale, contrast normalize, upscale small photos, sharpen, binarize) plus a receipt-oriented Tesseract page-segmentation mode, to improve recognition on phone photos and scans.
  • Raw OCR text panel in the receipt detail view (collapsible), so the exact recognized text is visible for review and correction.

Fixed

  • Date detection (EN + ES). Textual months are parsed in both languages and both orderings — e.g. 02 June, 2030, June 2, 2030, 02 de junio de 2030 — alongside numeric dates. Ambiguous numeric dates are disambiguated by range (US mm/dd/yyyy like 02/15/16 reads correctly), and unparseable/garbled dates leave the field empty and editable instead of showing NaN.
  • Amounts read from the correct column. Totals, subtotals and tax are extracted from the value column on the label's line, ignoring decoys such as a (3.8 %) tax rate; OCR-split cents (125 00 → 125.00) are repaired and a dropped-decimal tax larger than the subtotal is recovered (475 → 4.75).
  • English field keywords. TAX / VAT / GST and TOTAL DUE / AMOUNT DUE / BALANCE DUE / GRAND TOTAL are recognized alongside the Spanish forms; RFC (tax id) and UUID are extracted from OCR/plain text; the confidence score reflects how many fields were recovered.

Changed

  • Dependabot cadence reduced to quarterly (every 3 months) to cut PR noise.

Full changelog: https://github.com/TastingRogue/HomeLedger/blob/main/CHANGELOG.md · Compare: v1.1.0...v1.1.1