Top Handwriting OCR Tools for Precise Text Extraction

Does OCR work on handwriting?

Unlike printed documents, handwritten text recognition poses many challenges: lack of standardization, varied writing styles, unequal scan quality... However, with the advances of AI, some solutions now go well beyond simple automatic reading. In this article, we analyze the real level of reliability of OCR technologies on handwriting, through concrete cases, comparative tests and business use feedback.

Our test format

For this article, all solutions were tested using the exact same documents. We used one document from the IAM dataset for single-line handwriting recognition, one document from the FUNSD dataset for handwritten text inside structured forms, and finally a standard US invoice template containing a handwritten signature.

Test documents

Pen to Print

Test on Single Handwritten Line

On the single handwritten line test, Pen to Print performs correctly. No visible errors appear in the transcription.

Test on Handwritten Form

For the handwritten form, the text was correctly recognized and extracted. However, each detected text block is returned as a separate line. In the highlighted output, we can clearly see the areas of interest. Pen to Print does not distinguish between printed text and handwritten text in the rendering. The output remains quite raw and structured only with line breaks.

Test on Handwritten Invoice signature / real business document

As in test 2, the handwritten data was correctly extracted. You simply need to locate the line stating “Signature: dave” at the bottom of the output.

Result of the 3 tests

✅ No errors detected
‍To conclude on Pen to Print, it is a very effective solution for detecting and extracting handwritten text, even in more complex documents mixing handwritten and printed content. However, the output format remains quite basic and minimally structured. Retrieving specific data can therefore sometimes be time-consuming.

TrOCR

Test on Single Handwritten Line

For the handwritten sentence, TrOCR extracted the text perfectly, down to the smallest detail. Test 1 is therefore a success.

Test on Handwritten Form

Here, the output speaks for itself. TrOCR did not manage to properly read the document or at least did not understand its structure. It returned “ITV,” which is likely just a fragment of the original text. The model struggles with structured documents containing handwritten fields.

Test on Handwritten Invoice signature / real business document

TrOCR did not reliably transcribe the handwritten signature. Instead of extracting a name, the model produced non-alphabetic characters. This suggests the signature was treated more as a visual shape than readable handwriting.

Result of the 3 tests

❌ 2 errors detected
Because it does not include a contextual understanding layer like some other OCR software, TrOCR failed in 2 out of 3 tests. On pure handwritten line recognition, it performs very well. However, on stylized handwriting such as signatures and on structured business documents, it struggles. It remains a good solution for raw handwriting transcription.

OpenAI

Test on Single Handwritten Line

On this test, OpenAI extracted all the text but introduced a few minor errors. It correctly detected quotation marks in the sentence but added extra ones where there were none. The handwriting recognition worked, but the formatting slightly changed.

Test on Handwritten Form

For handwritten text extraction inside the form, OpenAI Vision performed well. The text was extracted without errors.

Test on Handwritten Invoice signature / real business document

As expected, the handwritten signature was correctly detected and transcribed. No errors appeared at this stage.

Result of the 3 tests

❌ 1 error detected
ChatGPT performed very well overall but made one formatting mistake in test 1. The hallucinated quotation marks did not prevent the extraction of the handwritten text itself, but they slightly altered the output. For the other tests, OpenAI delivered strong and reliable results.

Gemini

Test on Single Handwritten Line

Gemini successfully passed test 1. The handwritten text was fully extracted with all the correct information, including subtle elements such as quotation marks.

Test on Handwritten Form

Test 2 was perfectly executed. All handwritten fields were extracted correctly without any mistakes.

Test on Handwritten Invoice signature / real business document

Test 3 was also completed successfully. Gemini correctly detected and extracted the handwritten signature name.

Result of the 3 tests

✅ No errors detected‍
Gemini successfully passed all three tests without errors. Its integrated AI comprehension layer enables intelligent and context-aware handwriting recognition. It is a strong solution for this use case.

Mindee

Test on Single Handwritten Line

On the handwritten line test, Mindee correctly extracted the data. No errors were identified.

Test on Handwritten Form

On more complex documents, Mindee recognizes and extracts handwritten text within a defined extraction field.

Test on Handwritten Invoice signature / real business document

Result of the 3 tests

✅ No errors detected
For all three tests, I had to create a specific model dedicated to extracting handwritten text only. Once configured, Mindee correctly identified the relevant areas and extracted the information. It is therefore a strong solution, but you need to set up and configure your extraction template beforehand.

Koncile

Test on Single Handwritten Line

For the standalone handwritten line, Koncile’s handwriting recognition works correctly. The text was extracted with all its subtleties and without errors.

Test on Handwritten Form

Everything was correctly extracted here as well. In addition to the handwritten names, the associated dates were also extracted automatically. This test is therefore conclusive.

Test on Handwritten Invoice signature / real business document

The handwritten signature name was correctly extracted within the dedicated “handwritten text” extraction field.

Overall score of the 3 tests

✅ No errors detected
Across all three tests, the overall performance is highly satisfactory. No errors were identified. The app also allows you to extract additional complementary information, especially in tests 2 and 3. Extraction fields are generated intelligently based on the information detected in the document. There are specific fields dedicated to handwritten text recognition, as well as structured fields for the rest of the document data.

This structured approach makes data classification and business integration much easier.

Koncile is certainly one of the most advanced OCR solutions for handwritten text extraction.

Comparative table of results: precision, speed, business context

This comparison highlights the real differences between handwriting OCR tools when you move from simple transcription to real business documents. Some solutions are strong on isolated handwritten lines but struggle as soon as the layout becomes structured (forms, mixed printed and handwritten zones, signatures).

Others go further thanks to an AI understanding layer, which makes the output more usable in practice: better handling of handwriting variations, better extraction inside forms, and more reliable detection of signatures as actual names rather than shapes.

The gap is not only about raw accuracy, it is also about how the result is delivered.

Tools like Pen to Print can extract handwritten text, but the output remains raw and line-based, which makes it harder to retrieve one specific piece of information quickly.

On the opposite side, solutions that can structure results and isolate handwritten fields (Gemini, OpenAI Vision, Koncile, and Mindee once configured) are much more operational in real workflows. This is where intelligent document processing really matters: not just reading handwriting, but turning it into structured, actionable data.

Which OCR should you choose according to your needs?

Among all the solutions tested, Koncile stands out as the most complete and operational OCR for handwriting extraction. Its integrated AI does more than read text — it understands the document, structures the information, and allows model customization based on your specific use case. This makes it immediately usable in real business workflows.

Other tools serve different purposes depending on what you need:

• Pen to Print is effective for quick handwritten transcription, but the output remains largely unstructured.
• TrOCR performs very well on isolated handwritten text but struggles with complex document layouts.
• OpenAI Vision and Gemini are highly adaptable and accurate thanks to multimodal AI, but they function more as powerful technology components than full document processing systems.
• Mindee delivers strong results for structured extraction, but requires building and configuring a dedicated model beforehand.

In practice, the right choice depends on how you plan to use handwritten data.

For simple transcription, several tools perform well.

For structured, reliable, and business-ready extraction, Koncile remains the most advanced and operational solution.