Skip to content

PDF and archivingGuide

7 min read

How to make the text in a web page's images searchable

In a vector PDF of a web page the page's own text is already text; only pictures, canvases and video frames need recognizing. Here is how to find which parts need it, and four ways to do it, from macOS Live Text to Acrobat, OCRmyPDF and Page Scanner.

A saved web page has two kinds of text in it. The first is the page's own: headings, paragraphs, table cells, the words the browser set in a font. If the PDF was made from the page's layout rather than from screenshots, that text is still text, and you can already search, select and copy it. The second kind is text that is part of a picture: a screenshot inside an article, the labels a chart or a map draws into a <canvas>, a slide in a paused video. No PDF can search that until something reads the picture and lays the words over it as invisible text, which is what optical character recognition (OCR) does. So the first step is to find out which kind you have.

First, find out what needs reading

Open the PDF and search for a word from the body text. If it is found, the page's text is text, and only the pictures are left. If nothing is found anywhere, the whole file is a picture: a capture that scrolled the page and stitched screenshots together gives you one large image, and every word on it needs recognizing.

A PDF made by Page Scanner open in Chrome's PDF viewer, with two paragraphs of the page's text selected

Page Scanner says which one you have before you export. Its normal capture is vector, and the Output Format panel's PDF line says the text stays selectable. When it had to fall back to stitching screenshots, most often because DevTools was open on the tab, that line says instead that the PDF holds an image. Closing DevTools and scanning again gets you real text, which beats any recognition. How to save a web page as a PDF compares what the common ways of saving keep.

Method 1: Live Text on a Mac, to read one picture

If you need the words from one image once, you may not need a searchable file at all. On a Mac, Live Text works in Preview and in Safari: drag across the text in a picture, Control-click, and copy it, look it up, translate it or search the web for it. Apple's guides describe using the text there and then; they do not describe saving it into a file. And "Live Text isn't available in all regions or languages", with the list on Apple's feature-availability page.

Verdict: the fastest way to get a sentence out of a screenshot, and nothing to install.

Method 2: Adobe Acrobat's Scan & OCR

Acrobat on the desktop does this for a whole PDF. Adobe's steps: open the PDF, choose All tools, then Scan & OCR, then In this file, pick the page range and the language, and choose Recognize Text. "Acrobat creates a searchable text layer in your PDF." In multiple files does a batch. Adobe's plan comparison lists "Turn scanned documents into editable, searchable PDFs" under Acrobat Pro, at "US$19.99/mo", annual, billed monthly.

There is also a free online version: "Drag and drop a PDF file to recognize text in it using OCR technology, creating a searchable PDF." You sign in to download the result, and "Your file will be securely handled by Adobe servers and deleted unless you sign in to save it." That is fine for a public page and worth a thought for one you had to sign in to see.

Verdict: the one to use if you already pay for Acrobat Pro, or want correction tools after the recognition.

Method 3: OCRmyPDF, free and on your own computer

OCRmyPDF "adds an OCR text layer to scanned PDF files, allowing them to be searched or copy-pasted". It is open source, runs on Linux, Windows, macOS and FreeBSD, uses Tesseract "to recognize more than 100 languages", and writes PDF/A by default. It is a command, which suits a folder of files better than a single one.

What matters for a web page is how it treats a PDF that already has text. By default, "If a page in a PDF seems to have text, by default OCRmyPDF will exit without modifying the PDF", which is every page of a vector capture. --mode force rasterizes every page first, which turns the real text into a picture of text; you do not want that here. --mode redo is the one: its documentation says "If a file contains a mix of text and bitmap images that contain text, OCRmyPDF will locate the additional text in images without disrupting the existing text." A vector capture with a screenshot in the middle of an article is exactly that mix.

brew install ocrmypdfocrmypdf -l eng --mode redo page.pdf page-searchable.pdf

--mode arrived in OCRmyPDF 17; before it, the same thing is --redo-ocr.

Verdict: the free choice for any language Tesseract reads, and the one whose documentation covers reading the pictures inside a PDF that already has text.

Method 4: Page Scanner's Text in pictures

Page Scanner does the recognition while it makes the PDF. When the crop holds text that is a picture, the Output Format panel shows Text in pictures in its PDF group, with two choices: Recognize it, on this computer, the default, and Leave it a picture. At Extract it reads that text and lays it over the picture invisibly, so the PDF can be searched and the words selected and copied. The message after the export counts the words it laid in.

It reads three kinds of picture:

  • the <canvas> elements a map, a chart or a document viewer draws its text into;
  • the whole of a capture that fell back to stitched screenshots;
  • a picture, or a scanned PDF, opened from disk with Open a PDF or image, so a screenshot you already have can come out as a searchable PDF too.

The engine is Tesseract 5 compiled to WebAssembly, with models for English, Korean, Japanese, Simplified Chinese and Italian, all inside the extension. A page is read in its own language, taken from the page's lang, the browser's language and English, at most three. Nothing is downloaded, and the pictures are read on your computer and never leave it; it takes a few seconds for a canvas and longer for a whole page.

What it costs, and what it does not do:

  • Size. The engine and the five languages add about 11 MB to the extension's download.
  • Languages. Those five. Anything else is read as English.
  • Accuracy. Recognition is only as good as the capture's preview, so a very tall page and small or stylized type read worse, and Korean words can come out split into syllables. A word the engine is less than 45 % sure of is left out. The panel says that the text is recognized, can have mistakes and is not the page's own.
  • What it skips. Text in an <img> or a CSS background on a vector capture is not read, only canvases and whole-picture captures, and nor is a video's still. For the photos in an article, run the PDF through OCRmyPDF's --mode redo afterwards.
  • Only a PDF carries it. A PNG or JPEG export is the picture, and the Markdown and Word exports leave canvas text out. An agent's scan, from the command line or the MCP server, has no recognized text.
  • Marks. Nothing recognized under a blur, a redaction or an erase goes into the file.

If the engine does not start within 30 seconds, or fails, the PDF is made without recognized text and the editor says so; the export itself never fails over it. The capture, the recognized words and the file stay on your computer. The extension does send anonymous counts of which features are used, including whether recognition was on, never the words; the settings can switch that off.

Verdict: the one to use when the picture is a chart or map on a page you are saving anyway, in one of its five languages.

Which should you use?

Use Live Text if you want the words from one picture, now, and do not need a file.

Use Acrobat if you have Acrobat Pro and want a guided correction pass over the result.

Use OCRmyPDF if you are comfortable in a terminal, the language is not one of Page Scanner's five, or the text is in the photos of a vector PDF.

Use Page Scanner if the text is in a chart, a map or a viewer on the page you are capturing, or the whole capture is a picture, and you want it searchable in the same export, without the file leaving your machine.

Frequently asked questions

Why is none of the text in my PDF searchable? The PDF is one picture. With Page Scanner that means the capture fell back to screenshots, usually because DevTools was open on the tab; close it and scan again, and the page's text comes out as text. Otherwise, any of the four methods above will recognize it.

Is recognized text good enough to quote? Check it against the picture first. Every method here guesses from pixels, and all of them can misread a character, especially in small type.

Does recognition change how the PDF looks? Page Scanner's does not: the words go in as invisible text over the picture, so the page looks the same and the text is what a search, a selection or a copy finds. OCRmyPDF "Places OCR text accurately below the image", and Adobe says Acrobat "creates a searchable text layer".

Where this comes from

Keep reading