ocr-text-extract

Browser OCR with smart preprocessing, multi-pass recognition, and layout reconstruction — powered by Tesseract.js.

All processing runs in the browser. Images are never sent to a server.

Features

• Multi-pass OCR with automatic best-result selection
• Smart image preprocessing (contrast stretch, sharpening, adaptive threshold)
• Modal/dialog auto-focus for UI screenshots
• Multi-column layout reconstruction
• Clean mode with confidence-based word filtering
• Dark image and logo/single-line fallbacks
• 12+ language support via Tesseract

Install

npm install ocr-text-extract tesseract.js

tesseract.js is a peer dependency — install it alongside this package.

Usage

import { extractTextFromImage } from "ocr-text-extract";

// From a File input
const file = document.querySelector("input[type=file]").files[0];

const result = await extractTextFromImage(file, {
  language: "eng",
  cleanMode: true,
  output: "text", // or "json"
  onProgress: (pct) => console.log(`${pct}%`),
});

console.log(result.text);        // always plain extracted text
console.log(result.formatted);   // text or JSON string, based on `output`
console.log(result.output);      // "text" or "json"
console.log(result.confidence);  // 0–100 score, or null if empty
console.log(result.passUsed);    // e.g. "enhanced", "focused", "binary"

Output formats

| output | result.formatted | result.text | |----------|-------------------|---------------| | "text" (default) | Plain extracted text | Same plain text | | "json" | JSON string with text, confidence, passUsed, metrics | Plain text (always available) |

// JSON output example
const result = await extractTextFromImage(file, { output: "json" });
const data = JSON.parse(result.formatted);
// { text: "...", confidence: 87, passUsed: "enhanced", metrics: { ... } }

Input types

extractTextFromImage accepts:

• Data URL string ("data:image/png;base64,...")
• File or Blob

Options

| Option | Default | Description | |--------|---------|-------------| | language | "eng" | Tesseract language code | | cleanMode | false | Filter low-confidence words; best for screenshots | | targetLongEdge | 2000 | Target image long edge when scaling | | maxLongEdge | 3200 | Maximum long edge when scaling | | ocrDpi | "300" | DPI hint passed to Tesseract | | output | "text" | Result format: "text" or "json" | | onProgress | — | Callback receiving 0–100 progress |

Lower-level API

Individual preprocessing and formatting helpers are exported for advanced use:

import {
  loadScaledCanvas,
  preprocessFromScaled,
  detectModalRegion,
  formatOcrOutput,
  normalizeText,
  scoreOcrResult,
  SUPPORTED_LANGUAGES,
} from "ocr-text-extract";

Browser support

Requires a modern browser with canvas, Image, and FileReader APIs. Designed for client-side use only.

Development

npm install
npm test
npm run build

License

MIT