Marathi OCR — image to text

Marathi · measured on this model

What the Marathi model got wrong in our test.

Marathi shares the Devanagari script with Hindi but adds letters and conjuncts of its own, notably ळ and the eyelash ऱ्य, and has its own model. In our test the Marathi model made no character errors in five of the seven renders; the errors we did see were mostly in the letters that distinguish Marathi spelling, and they came with faded print or a poor photo. The checklist lists those first.

How we tested: we wrote 12 lines of Marathi (360 characters, none copied from a third party), drew them in two Noto fonts (SIL OFL), degraded them in 7 ways and ran the same recognition engine and Marathi file this page uses, with its default settings. We aligned the output to the text letter by letter. This is a small synthetic test that shows real failure modes. It is not an accuracy figure for your documents, and real scans and photos will behave differently. A script in the source code reproduces it byte for byte; the exact versions are under Sources at the end of this page.

Character errors by condition

• Clean render, sans, 36 px

  0.0% character errors

  Engine confidence 95

• Clean render, serif, 36 px

  0.0% character errors

  Engine confidence 95

• Small text, sans, 18 px

  0.0% character errors

  Engine confidence 92

• Photo-like: sans 28 px, 1.5° tilt, blur, noise, JPEG q45

  0.0% character errors

  Engine confidence 95

• Tiny text, sans, 13 px

  0.0% character errors

  Engine confidence 89

• Poor phone photo: sans 22 px, 3° tilt, blur σ1.4, heavy noise, JPEG q30

  9.2% character errors

  Engine confidence 91 · 1 of 12 lines omitted

• Faded print: serif 20 px, grey ink on tinted paper, blur σ1.1

  2.8% character errors

  Engine confidence 91

Proofreading checklist

Marathi: check these before you trust the text.

Items marked “observed” come from the test above, with the number of the 7 conditions in which each one appeared. Items marked “look-alikes” are a general proofreading habit for this script that we did not measure.

1. ऱ्य र्य

   Observed · 1 of 7 conditions

   01The eyelash ra

   In faded print the model read the eyelash ऱ्य in शेतकऱ्यांना as a plain र्य (शेतकर्यांना), which is a different spelling in Marathi. Search your result for र्य and compare each one with the source.

   • शेतकऱ्यांना → was read as शेतकर्यांना

2. ज्ञ ज

   Observed · 1 of 7 conditions

   02ज्ञ at the start of a word

   In a degraded photo ज्ञानेश्वरांची was read as जनेश्वरांची: the conjunct ज्ञ and its aa sign collapsed into ज. The word looks plausible, so only a comparison with the original will catch it.

   • ज्ञानेश्वरांची → was read as जनेश्वरांची

3. वि दि

   Observed · 1 of 7 conditions

   03व with an i-sign read as द

   The first letter of विद्यार्थ्यांची was read as द (दिद्यार्थ्यांची) in the same degraded photo. Names and headings are the most visible place for this.

   • विद्यार्थ्यांची → was read as दिद्यार्थ्यांची

4. रू रु

   Observed · 1 of 7 conditions

   04The long u sign

   In faded print भरून lost its ू (भरन). Long and short u signs, and the signs under the line in general, are the first to vanish when ink is pale.

   • भरून → was read as भरन

5. । ) !

   Observed · 1 of 7 conditions

   05Sentence-end marks and stray punctuation

   The full stop danda was read as ) in one degraded photo and as ! in faded print, and a stray apostrophe and full stops were added before or after words. Delete punctuation that was not in the original.

6. ळ ल ण न

   Look-alikes · not measured

   06ळ and ण

   The Marathi-specific letters ळ and ण resemble ल and न. In our test, ळ was read correctly in every word and every condition except on the line the model omitted entirely, but small print and your own typeface may behave differently. Re-read place names such as कोकण and कोल्हापूर.

Photographing Marathi

Tips for photographing Marathi documents.

• Choose the Marathi model, not Hindi

  The Hindi and Marathi models are separate files. Marathi text with ळ and ऱ is better read with the Marathi model; the model dropdown on this page selects it.

• Watch for whole lines going missing

  In our poorest photo condition the model omitted a whole line (शेतकऱ्यांना कर्जमाफी मिळणार आहे) without any warning. Count the lines in the result against the original, especially in forms and notices.

• Keep the headline straight and the top margin clear

  As with Hindi, the shirorekha joins the letters of a word. Rotate the photo so headlines are level, and leave a margin above the first line so reph and the signs above the line are not cropped.

• Prefer the original over a photo of a photo

  Notices in Marathi are often photocopied. The fainter and blurrier the copy, the more the long u and the eyelash ra are lost; scan or photograph the best copy you have, flat and evenly lit.

• Check the confidence number against the page

  In the degraded photo above, the engine’s average confidence stayed around 91 while 9% of characters were wrong. Do not treat the confidence figure as the accuracy; proofread.

General advice on resolution, binarisation, skew and borders is in the Tesseract documentation.

Legacy-font PDFs

Marathi PDFs that need the PDF tool, not OCR.

Marathi newspapers, forms and government circulars were often typed in legacy fonts such as Shree-Dev 0714, Shivaji or Kiran and exported to PDF. Those PDFs usually carry a Latin-letter text layer. When the font name survives in the file, the PDF tool can decode that layer directly and avoid the recognition errors listed above.

Use the PDF to Unicode tool when the PDF has a text layer. Use OCR for a scan or a photo, which has no text layer, and when the PDF tool says it cannot recognise the font.

3 Marathi legacy fonts the engine converts

• Shree-Dev 0714
• Shivaji
• Kiran

Model download

The Marathi recognition file: about 2.1 MB, downloaded on first use.

Recognition runs in your browser. The first time you press Recognise, the Marathi recognition file downloads from a public file host (jsDelivr); your image is never sent there. The English file, for comparison, is about 4.1 MB. Your browser normally keeps the file, so later visits on the same browser skip the download; clearing site data removes it.