Scanned Document to Text: How to OCR Phone Scans and PDFs

Get a readable scan first

OCR, optical character recognition, can only read what the camera caught. Docubox's notes on the tool say that accuracy depends on scan quality, resolution, skew and contrast. A few habits help.

• Flatten the page. Lay it on a table and hold down curled corners, because a curve bends the lines of text.
• Light it evenly. Keep the shadow of your hand and phone off the page, and watch for a bright glare spot on glossy paper.
• Shoot from straight above. Text photographed at an angle comes out slanted and narrower at one end.
• Keep it sharp and close. Let the page fill most of the frame and tap to focus before you take the picture.
• Make sure it is upright. Docubox applies the rotation flag that phone photos carry, but it does not detect text that is sideways or upside down in the image, so turn such a picture before you upload it.

Your phone's scanner handles much of this for you. In Notes on an iPhone, Scan Documents captures the page automatically and lets you drag the corners to fit it (How to scan documents on your iPhone or iPad). On Android, the Google Drive app outlines the page, crops to that outline and saves the scan as a PDF or JPG (Scan documents with Google Drive).

Turn the scan into text

1. Add the scan. Open OCR to text and add a JPG, PNG or WebP image or a PDF. The upload area states both conditions: processing happens on Docubox's servers, and downloading needs a free account.
2. Check the Layout option. The options appear under the file. Language offers English only. Layout starts on Automatic, which lets the engine find the blocks of text on the page; Single column treats the whole page as one block and is worth trying on a plain letter if Automatic splits lines oddly.
3. Select Convert to TXT. The file uploads and Docubox reads it one page at a time; the progress bar counts finished pages.
4. Sign in to get the text. When the run finishes, sign in with your Google account or an emailed code, free and without a password, and download the .txt file.
5. Keep it if you need it. Choose Save to my documents for a lasting copy; an unsaved text file is removed when its time runs out.

If your document is a set of separate photos, one per page, combine them with Images to PDF first, so the pages go through OCR together and come back in one file. PDF pages are read at a fixed resolution, so for very small print, run the photo on its own.

Reading the result

You get a plain .txt file. Each page opens with a separator line that gives the page number and a confidence percentage, followed by the text found on that page. Every line of text keeps its own line, with a blank line wherever the engine sees a new paragraph or block.

The confidence figure is the average of the engine's own scores for the words it found on that page. It says how sure the engine was, not how many words are right, and a page with no words found shows zero. Pages that fall below Docubox's threshold also get a warning on the result page asking you to review their text.

Proofread against the original, starting with the flagged pages. Check figures, names, dates and reference numbers character by character, since those are the details that are costly to get wrong.

What OCR can't do here

• Other languages. The engine is set to English. Text in other languages, and in other alphabets, is likely to be missed or come out wrong.
• Handwriting. Only printed text is recognized, so handwritten notes in a margin will be missing or garbled.
• Layout. Columns, tables, fonts and formatting are not kept. A table comes out as lines of words, not rows and cells.
• Searchable PDFs. The result is a separate text file. Your PDF is not changed, and no text layer is added to it.

For a few lines from a photo, your phone may be quicker. On an iPhone that supports Live Text, you can touch and hold a word in a photo and copy the text from there (Copy and translate text from photos).

Already has text? Skip OCR

A PDF saved from a word processor or other software carries its text inside the file, and so does a scan that a scanner app has already run through OCR. Try selecting a sentence in your PDF viewer. If it highlights, use PDF to text instead: it copies the stored text rather than reading the page again, and it does not use your OCR pages. One quirk: the reading order follows the way the PDF stores its text, which is not always the order you see on the page.

An image counts as one page. Each run also counts as one conversion. When a run starts, Docubox sets aside pages from your allowance and afterwards counts only the pages it read, but a PDF with more pages than you have left is refused rather than read in part. For a long PDF, sign in before you upload it.

What happens to your files

Docubox runs the recognition on its servers, not on your phone or computer. The scan stays in private storage until it expires, and Docubox's systems can read it while they work on it. If a page is too sensitive to upload anywhere, retype it or use software that runs entirely on your own device.

Only the text file can be saved; the uploaded scan always expires. Downloading leaves the file in place and does not reset the clock. The retention policy explains the rest.