NDLOCR-Lite Region-Select OCR (Chrome Extension)

日本語版はこちら

A Chrome extension that runs fully offline, in-browser Japanese OCR using ndlocr-lite-wasm. Select a region on any page with your mouse, and the recognized text is automatically copied to your clipboard. No data ever leaves your browser.

Features

• Launch from the toolbar icon, Alt+Shift+O, or the right-click context menu
• Drag to select a region (Esc to cancel)
• DEIM text-region detection → XY-Cut reading order → PARSeq character recognition (up to 4 lines recognized in parallel)
• Results are auto-copied via navigator.clipboard.writeText with a toast notification
• Models are cached in IndexedDB after the first load for fast subsequent runs
• Image viewer: open local images or clipboard images for OCR (zoom + region select supported)
• PDF viewer: open a local PDF and OCR any page you like via page navigation (password-protected PDFs not supported)
• Deskew & Select: fit four corner handles to a page photographed at an angle; it's flattened before OCR (viewer)
• Eraser: paint over stamps, handwritten notes or anything else you don't want OCR'd before running OCR, with Ctrl+Z undo (viewer)
• OCR history: results are automatically saved to chrome.storage.local for later re-copy, edit, search, and delete
• Cleaning rules: customize regex-based post-processing rules
• Same-row detection for tables/receipts produces tab-separated output
• Automatic furigana (ruby) filtering
• Robust against tight selections, tiny images, and captures with adjacent-line fragments (v0.2 / v0.6 / v0.7.1)

Build

npm install
npm run build

The extension files are output to dist/.

Load the Extension

1. Open chrome://extensions/ in Chrome
2. Enable Developer mode (top right)
3. Click Load unpacked and select the dist/ folder

Usage

1. Open a page containing the text you want to OCR
2. Click the extension icon in the toolbar → Start region selection (or press Alt+Shift+O, or right-click → Region-Select OCR (NDLOCR-Lite))
3. Drag to select the region
4. Wait a few seconds — progress toasts will appear, ending with Copied (N chars)
5. Paste anywhere

Architecture

flowchart LR
  subgraph Content["Content Script"]
    CS_UI["Selection UI / Rectangle"]
    CS_Toast["Progress Toast"]
    CS_Clip["Clipboard Write"]
  end

  subgraph Background["Background Service Worker"]
    BG_Cmd["Command Receiver\n(icon / shortcut / menu)"]
    BG_Cap["captureVisibleTab"]
    BG_Route["Message Routing"]
  end

  subgraph Offscreen["Offscreen Document"]
    OS_Crop["Image Crop\n(OffscreenCanvas)"]
    subgraph Worker["OCR Web Worker"]
      W_DEIM["DEIM Detection"]
      W_Parse["NDL Parser\n+ XY-Cut Reading Order"]
      W_PARSeq["PARSeq Recognition"]
    end
  end

  BG_Cmd -->|"scripting.executeScript\n+ start-selection"| CS_UI
  CS_UI -->|"selection rect + dpr"| BG_Cap
  BG_Cap -->|"screenshot dataURL"| OS_Crop
  OS_Crop --> W_DEIM --> W_Parse --> W_PARSeq
  W_PARSeq -->|"recognized text"| BG_Route
  BG_Route -->|"ocr-result"| CS_Toast
  CS_Toast --> CS_Clip

Loading

Context	Source	Role
Content Script	src/content/content.ts	Selection overlay, progress toast, clipboard write
Background SW	src/background/service-worker.ts	Command aggregation, captureVisibleTab screenshot, forward to offscreen, return result to content
Offscreen Document	src/offscreen/offscreen.ts	ONNX Runtime Web cannot run inside a SW, so an offscreen document hosts the OCR Worker. Models are preloaded on startup
OCR Worker	src/ocr/worker/ocr.worker.ts	Ported from ndlocr-lite-wasm. Runs DEIM (detection) and PARSeq (recognition) via onnxruntime-web/wasm

File Size

Component	Size
DEIM fp32 model	38 MB
PARSeq fp32 model	39 MB
ONNX Runtime WASM	12 MB
JS / HTML / icons	< 1 MB
Total (dist/)	~90 MB

Well within the Chrome Web Store upload limit (2 GB).

Acknowledgements

The OCR engine and models in this extension are based on NDLOCR, researched, developed, and published by the National Diet Library of Japan (NDL). We deeply appreciate the NDL for openly sharing their high-accuracy Japanese OCR technology.

The lightweight web port ndlocr-lite-wasm serves as the foundation of this extension's OCR pipeline.

Changelog

See CHANGELOG.md for the version history.

License

• This repository: CC-BY-4.0 (inherited from ndlocr-lite-wasm)
• Model files: subject to the NDLOCR (National Diet Library) license
• ONNX Runtime Web: MIT