How to change text in a scanned PDF without reflowing the page

Steps

1. Open the fileDrag the scanned PDF into the window. Encrypted files show a password box, and "Keep original password" is on by default. Recognition of scanned pages starts in the background right away.
2. Enter keywordsIn "Single" mode, fill in "Find text" and "Replace with". To change many items, switch to "Batch" and import a keyword table.
3. Check recognition language and fontLeave "Recognition language" on "Auto". "Replacement font" defaults to "Auto", which picks an installed font for the script and matches weight and darkness to the original ink.
4. Find and reviewClick "Find". Results appear while recognition runs. Click a row and the preview on the right jumps to that page and outlines the match. Entries with confidence below 80% have a warm-colored background; check those carefully.
5. Preview the replacementSwitch the preview to "Preview replacement" to see the current page after replacement. Dashed boxes mark the changes. This step does not save any file.
6. Replace and save asClick "Replace selected and save as…". The original file is not overwritten. On save, each change is checked in the new file by read-back verification, and the result appears in the "Read-back" column.

Why the usual methods reflow the page

Most PDF editors handle a scan by first recognizing the whole page as text, then letting you edit it as you would in Word. The recognized text uses substitute fonts, and the software recalculates line spacing and line breaks, so the page no longer matches the original. Another approach is to cover the text with a white box and type over it. When the background is not pure white, this leaves a visible patch.

In-place correction changes only the target text. For how it works, see Whole-page OCR reflow vs in-place correction.

10 common mistakes

1. Saving without previewing: a replacement longer than the original is written at the original font size by default, and may run into the text on its right.
2. Ignoring “Needs review”: check every match with confidence below 80% and every loose match.
3. Wrong recognition language: a Thai keyword will not be found with the Chinese recognition model. “Auto” is the safest choice.
4. Mixing full-width and half-width characters: image text matching treats full-width letters and digits as half-width, but text layer matching does not.
5. Letter case: text layer matching is exact and case-sensitive. Turn on “Loose match” to ignore differences such as 0/O, 1/l and case.
6. Text too small or resolution too low: text below about 150 DPI, or very small characters, is recognized poorly. If you can rescan, use 300 DPI.
7. Text over a stamp or watermark: the repair is less clean. Zoom in on the preview to check.
8. Words split across lines: only single-line text is supported. Split such a word into two keywords.
9. Chinese and Thai in one replacement: this needs a font that contains both scripts. If none is installed, you are asked to replace them separately.
10. Replacing text in a signed document: the software asks for confirmation or refuses, according to its risk levels. Such documents should usually only be redacted.

FAQ

Does it work on photos of documents?

Yes, if the text lines are roughly level, with skew within about 3°. Photos with strong perspective or curved paper are not supported yet.

Is the file larger after editing?

Almost the same. Edited images are saved again in their original format, using the original JPEG quantization tables.