Replace A PDF's Text Layer

Scanned PDFs often carry an old or poor OCR text layer. To add your own, first remove the existing text from every page, then write the new words at the positions your OCR engine reports.

var Recipe = require("@muhammara/native").Recipe;

function replaceTextLayer(inputPath, outputPath, words) {
  var recipe = new Recipe(inputPath, outputPath);
  var pages = recipe.read().pages;

  for (var pageNumber = 1; pageNumber <= pages; pageNumber++) {
    recipe.removeText(pageNumber, { forms: true });
  }

  words.forEach(function (word) {
    recipe
      .editPage(word.page)
      .text(word.text, word.x, word.y, { size: word.size, opacity: 0 })
      .endPage();
  });
  recipe.endPDF();
}

removeText() drops the text-showing operators (Tj, TJ, ', ") and keeps everything else: the scanned image, vector graphics, and marked content. { forms: true } also cleans the Form XObjects each page paints. OCR tools often put their text layer in one, and Recipe's own editPage() text lives there too. Leave it off when a form holds content that should keep its text, such as a letterhead shared with other pages.

Opacity 0 keeps the new words invisible but still searchable and selectable. Because text added with editPage() in the same Recipe is never removed, the order of the two steps does not matter.

Limits:

• Annotation appearances, such as filled-in form fields, keep their text.
• Content streams and forms are rewritten in place, so one shared with another page loses its text there too.
• The result is an incremental update. The old text bytes can remain in the file, so this is not redaction.

To change a few words and keep the page's own fonts instead, see Replace Text In An Existing PDF.