Extract Text from Images (OCR)

How to extract text from an image

1. 1

   Add the image

   Drop in a photo, screenshot, or scan containing text. The recognition engine works online, so the picture is never retained by PilotPDF — which matters when the thing you are reading is a document rather than a sign.

2. 2

   Run the recognition

   The image is analysed and the characters it contains are identified. Larger or noisier images take longer, because the work is being done by your own processor rather than a datacentre.

3. 3

   Copy the text or save a searchable PDF

   Review what came back and copy it, download it as a .txt file, or download a searchable PDF of the picture. Always proofread, because OCR is very good but not perfect.

What decides whether OCR works well

Optical character recognition is pattern matching against shapes, so the quality of the input governs the quality of the output far more than any setting does. Resolution is the first factor: text needs to be reasonably large in the image for the engine to resolve individual letters. A photo of a page taken from across a desk will produce noticeably worse results than the same page filling the frame, even though both are 'readable' to you.

Contrast and evenness of lighting come next. Crisp black text on white paper is the ideal case. A photograph with a shadow falling across half the page, a glossy page reflecting a window, or faded grey text on cream stock all give the engine less signal to work with. If you can retake the shot with flat, even light and the page filling the frame, that single change usually helps more than anything else.

Then there is the text itself. Clean printed type in a common font is recognised very reliably. Handwriting is substantially harder and often unusable. Decorative or condensed fonts, text over a photograph, tables and multi-column layouts, and pages photographed at an angle all reduce accuracy — skew in particular, because letters that are rotated or keystoned no longer match the shapes the engine expects.

Where OCR genuinely helps, and where to be careful

The everyday uses are the ones worth knowing: pulling a paragraph out of a screenshot instead of retyping it, capturing a serial number or Wi-Fi password from a photo of a label, lifting text from a slide or an infographic, getting a quotation out of a photographed book page, and turning a scanned form into something you can search.

Two habits are worth forming. First, always proofread the output. OCR mistakes cluster around characters that look alike — 0 and O, 1 and l and I, 5 and S, rn and m — and those are exactly the errors that slip past a quick glance in a reference number or a postcode, where they matter most. Second, expect layout to be lost. OCR returns characters, not design; columns, tables, and indentation usually come back as a linear run of text that needs restructuring.

If the source is a scanned PDF rather than a single image, PDF OCR is the better route, because it reads every page in one go and gives you all the text together.

Related tools