Image to JSON

How to convert Image to JSON

1. Add an image. A photo, scan, screenshot, or a table.
2. Pick a shape. Plain text, a word list with coordinates, or rows and columns.
3. Download the .json. Ready to parse in any language.

About Image to JSON

This is Image to Text for people who are going to write code against the result. Instead of a flat block of prose, you get structured JSON, and crucially, you get the metadata that a plain text export throws away: where on the page each word sat, and how confident the recogniser was about it.

Confidence scores are the part worth understanding, because they change how you can use OCR output. Recognition is never perfect, and the difference between a usable pipeline and an unreliable one is usually whether you can tell which parts to trust. A word returned at 96 percent confidence is almost certainly right. One at 45 percent needs review. With that number in the output you can route low-confidence fields to a human instead of silently importing a wrong digit into a spreadsheet.

Three output shapes cover the common cases. Plain gives you the full text with per-page confidence, which suits search indexing and archiving. Words gives every recognised word with its bounding box and confidence, which is what you want for building a form parser, redacting a document, or overlaying a text layer on an image. Table gives inferred rows and columns as arrays, matching what Image to Excel produces but in a format a script can consume directly.

Everything runs on your device via Tesseract compiled to WebAssembly. That matters more for a developer-facing tool than it first appears: the documents people build OCR pipelines against are usually invoices, receipts, forms, and identity documents belonging to their customers. Prototyping against a free web API means sending someone else's data to a third party, often without having thought about it. Here there is nothing to send.

Output shapes

OCR to JSON

If you came here looking for OCR to JSON, this is the tool. The difference from Image to Text is what comes back: instead of a flat block of text you get structured output. Every recognised word with its bounding box, its confidence score, and the line and paragraph it belongs to. That structure is what makes it useful in code. A script can filter by confidence, find the word nearest a coordinate, or rebuild the layout, none of which is possible from plain text.

Starting from a screenshot? PNG to JSON is the same output tuned for lossless PNGs. If you only need the table, PNG to CSV rebuilds the rows and columns for you. For the whole process, from capture to checking the result, see how to get a table out of a screenshot.

Image to JSON FAQ