Indian languages and handwriting · Indian Languages pack
Handwriting OCR test data: forms with handwritten fields and ground truth
To test handwriting OCR you need forms with handwritten entries and known correct values. GritWorks offers free synthetic Hindi, Bengali and Marathi forms whose handwritten entries (names, dates, amounts, addresses) come with JSON ground truth and bounding boxes, clean and degraded, so you can measure field-level accuracy instead of eyeballing a conversion.
On these forms the labels are printed and every annotated value is handwritten: names, dates of birth, amounts in figures and in words, addresses and signatures. Handwriting varies stroke by stroke, runs outside the box and crosses printed lines, so an engine that reads the printed form well can still miss most of the entries.
Field-level vs character-level scoring
Character error rate shows how close a reading is; field exact match shows whether the value is usable. Report both, per field type. A date with one wrong digit has a low character error and is still wrong, and comb-box fields such as phone numbers should be compared with whitespace removed.
What’s available now, and what’s next
The free sample holds five two-page forms: two Bengali, two Hindi and one Marathi, each with a degraded copy. English handwritten forms are not in the sample yet; if you need Latin-script handwriting, describe the forms on a call.
Degraded handwriting
Each form’s twin applies one condition: strong blur, coffee stain, photocopy, low light or cropped edges. Blur merges thin strokes, low light drops faint ink, and cropping cuts entries near the margin. Score each form beside its degraded copy to see which errors the condition added.
Handwritten clinical documents
Handwriting is also the hard part of clinical paperwork. The handwritten clinical documents case study describes how a healthcare organization turned restricted physician handwriting into a de-identified corpus for retrieval and agent evaluation.
FROM THE PACK’S GROUND-TRUTH JSON
Fields with ground truth
Every annotated field in the 5 documents, with one example value. Repeated rows such as line items are shown once. Identifier values are withheld from this page and stay in the pack.
फ्लैट नं. 407 साईं अपार्टमेंट गली नं. 4 शास्त्री नगर
Yes
present_address_line2
तलवंडी कोटा - 324005
Yes
permanent_address_line1
फ्लैट नं. 407 साईं अपार्टमेंट गली नं. 4 शास्त्री नगर
Yes
permanent_address_line2
तलवंडी
Yes
telephone
Withheld here · in the pack JSON
Yes
mobile
Withheld here · in the pack JSON
Yes
email
Withheld here · in the pack JSON
Yes
birth_cert_no
B/2021/07425
Yes
cert_issue_date
27/08/2022
Yes
issuing_authority
जन्म एवं मृत्यु पंजीकरण कार्यालय कोटा
Yes
id_proof
आधार कार्ड
Yes
address_proof
टेलीफोन बिल
Yes
specimen_signature
संदीप तिवारी
Yes
guardian_signature
संदीप तिवारी
Yes
declaration_date
06/09/2026
Yes
nominator_name
संदीप कुमार तिवारी
Yes
nominee_name_relation
प्रियंका देवी तिवारी (माता)
Yes
nominee_share
100%
Yes
SBI account opening form Part-I: Customer Information Sheet (CIF), Marathi27 annotated values · 2 pages · degraded copy: cropped edges
Annotated fields in marathi_01_clean.pdf
Field
Example value
Bounding box
form_date
17092026
Yes
branch_name
विश्रामबाग शाखा
Yes
branch_code
43524
Yes
customer_id
20383069840
Yes
name_title
श्रीमती
Yes
applicant_name
सुवर्णा संजय वाघ
Yes
date_of_birth
27011975
Yes
father_or_spouse_name
संजय शंकर वाघ
Yes
mother_name
पार्वती सुधाकर कांबळे
Yes
country
भारत
Yes
occupation
अभियंता
Yes
birth_city
सांगली
Yes
citizenship
भारतीय
Yes
pan
Withheld here · in the pack JSON
Yes
mobile
Withheld here · in the pack JSON
Yes
email
Withheld here · in the pack JSON
Yes
residence_telephone
Withheld here · in the pack JSON
Yes
id_document_no
FNBPW3987B
Yes
id_issue_date
14092001
Yes
address
घर क्र. 442 महात्मा फुले मार्ग
Yes
city
सांगली
Yes
district
सांगली
Yes
state
महाराष्ट्र
Yes
pincode
416416
Yes
applicant_signature
सुवर्णा वाघ
Yes
place
सांगली
Yes
signature_date
17092026
Yes
FREE SAMPLE · BY EMAIL
Get the Indian Languages sample pack
These documents ship in the Indian Languages pack: 10 PDFs, five clean and five degraded, with structured JSON and bounding boxes. Enter your name, email and company and we’ll email the link. No call or platform account is required for the sample.
The sample is a starting point. We can support the volume your project needs. Book a call to discuss your document types, languages, formats and target volume.
We also provide Excel spreadsheets and images in JPEG, PNG and TIFF formats. Qualified teams can receive the full PDF dataset free after the booking form and call.
The free offer covers PDFs. Larger volumes and other formats are scoped on the call. Product support is available around GritWorks products.
QUESTIONS
Handwriting OCR test data, answered.
Is this a handwriting-to-text converter?+
No. This page offers test data: forms with handwritten entries and the correct values, so you can measure how well a converter or OCR engine reads handwriting.
Which languages have handwritten samples?+
Hindi and Marathi (Devanagari script) and Bengali. Every annotated field on the five forms is a handwritten entry on a printed form.
How is handwriting ground truth created?+
The expected value of each field is recorded with the document, not read back with OCR, and each field has a bounding box in PDF points. Degraded copies keep the same expected values, even where an entry is obscured or clipped.
Can I train a handwriting model on this?+
Five forms are enough to prototype a pipeline and to build regression tests, not to train a handwriting model. Larger volumes are scoped on a call.
How do I get the sample?+
Choose “Email me this pack” and enter your name, email and company. We’ll email the download link. No call or platform account is required for the sample.
Can I use my own tools?+
Yes. Use the PDFs and reference files in your own evaluation, sandbox or testing workflow. The included reference data is not an automated scoring system. The sample pack is licensed CC BY 4.0: you may use, adapt and share it, including commercially, with attribution to GritWorks. Third-party names or logos that appear in some documents belong to their owners and are not covered by the license.