TorReader PDF 2.4.0 — OCR for scanned documents

TorReader PDF 2.4.0

New interface

The whole app moved to a high-contrast theme in both light and dark mode. Borders, hover and
selection states are consistent everywhere, the page sits on a neutral grey desk so its edge is
always visible, and the sidebar no longer competes with the toolbar for attention.

OCR for scanned documents — 10 languages

Scanned PDFs have no text layer, so you cannot search, select or copy anything in them.
TorReader PDF now recognises the text and makes it work like a normal document.

• Bundled, no download, no account, fully offline. The language data ships inside the app.
• 10 languages: Vietnamese, English, Chinese (Simplified), Chinese (Traditional), Japanese,
  Korean, French, Russian, Portuguese, Spanish. Vietnamese and English use the highest-accuracy
  models.
• Your file is never modified. The recognised text lives in memory only — the file on disk
  keeps the same checksum.
• A new OCR tab in the sidebar: recognise the whole document or just the current page, pick the
  language, watch progress, cancel any time.

Select, find and copy text — in PDFs and in scans

• Select text works by character, the way a text editor does: drag from the middle of one word
  to the middle of another, across lines and across pages. Double-click selects a word, triple-click
  a line, Ctrl+A the page.
• Find ignores diacritics by default, so typing MAT also finds MẶT — useful when a scan
  loses a mark here and there. Turn on Match diacritics for exact matching.
• Search results now scroll straight to the match instead of only jumping to the page, and
  highlights stay visible as you scroll through the document.
• Copy with Ctrl+C or right-click → Copy text.

Links

Links inside a PDF now work. Internal links jump to the target and briefly outline it; external
links ask for confirmation and show the full address before your browser opens.

Faster

Opening and flipping pages no longer waits on background work. A single shared page cache means a
page is parsed once instead of several times, and neighbouring pages are prepared while you read.
On a large CAD drawing the per-flip cost dropped from about 850 ms to 1–2 ms after the first
visit.

Fixed

• Comments landed in the wrong place on pages rotated 270°.
• Fill-only shapes were drawn with an extra outline.
• Search highlights disappeared when scrolling in continuous mode.
• Opening a second document kept the first document's search results in the sidebar.
• Licence text in the About dialog showed mojibake on Windows.

---

Windows: portable ZIP, no installer, no admin rights.
Linux: AppImage, .deb via the APT repository, Snap and Flatpak.