OCRResult

ocr.OCRResult

One element of the array returned by OCR.detect: recognized text, confidence, axis-aligned bounds, optional rotation / rotationConfidence, and a helper click() that taps the box center via accessibility.

Table of contents

Constructors

• constructor

Properties

• bounds
• confidence
• javaObject
• rotation
• rotationConfidence
• text

Methods

• click

Constructors

constructor

• new OCRResult(javaObject, bounds, text, confidence, rotation, rotationConfidence)

Internal constructor—normally you only receive instances from detect.

Parameters

Name	Type	Description
javaObject	JavaObject	Native OCR result handle.
bounds	Rect	Bounding box in image coordinates.
text	string	Recognized UTF-8 text.
confidence	number	Quality score 0…1.
rotation	number	Clockwise angle degrees when rotation detection is enabled.
rotationConfidence	number	Confidence for rotation, 0…1.

Properties

javaObject

• Readonly javaObject: JavaObject

Underlying JNI object. Rarely needed from script; some engines may expose extra fields (lines, tokens, etc.).

---

bounds

• Readonly bounds: Rect

Bounding quadrilateral projected to a Rect: left, top, right, bottom, plus derived centerX, centerY, width, height.

Field cheat sheet

• bounds.left — left edge X
• bounds.top — top edge Y
• bounds.right — right edge X
• bounds.bottom — bottom edge Y
• bounds.centerX, bounds.centerY — centroid (read-only accessors)
• bounds.width, bounds.height — span in pixels (read-only accessors)

Example

"nodejs";
const { createOCR } = require("ocr");
const { requestScreenCapture } = require("media_projection");

async function main() {
  const ocr = await createOCR();
  const capturer = await requestScreenCapture();
  const img = await capturer.nextImage();
  const results = await ocr.detect(img);

  for (const result of results) {
    const bounds = result.bounds;
    console.log(`text: ${result.text}`);
    console.log(`origin: (${bounds.left}, ${bounds.top})`);
    console.log(`size: ${bounds.width} x ${bounds.height}`);
    console.log(`center: (${bounds.centerX}, ${bounds.centerY})`);
    console.log(`corner: (${bounds.right}, ${bounds.bottom})`);
  }

  ocr.release();
  capturer.stop();
}
main();

---

text

• Readonly text: string

Recognized string for this region.

Example

"nodejs";
const { createOCR } = require("ocr");
const { readImage } = require("image");

async function main() {
  const ocr = await createOCR();
  const img = await readImage("./sample.png");
  const results = await ocr.detect(img);

  for (const result of results) {
    console.log(`recognized: ${result.text}`);
  }

  ocr.release();
}
main();

---

confidence

• Readonly confidence: number

Model confidence ∈ [0, 1]; higher is better.

Rough bands

• 0.9 – 1.0 — very likely correct
• 0.7 – 0.9 — usually reliable
• 0.5 – 0.7 — verify visually
• 0.0 – 0.5 — suspect / noisy

Example

"nodejs";
const { createOCR } = require("ocr");
const { requestScreenCapture } = require("media_projection");

async function main() {
  const ocr = await createOCR();
  const capturer = await requestScreenCapture();
  const img = await capturer.nextImage();
  const results = await ocr.detect(img);

  const highConfidenceResults = results.filter((r) => r.confidence > 0.8);
  console.log(`high-confidence count: ${highConfidenceResults.length}`);

  for (const result of highConfidenceResults) {
    console.log(
      `text: ${result.text}, confidence: ${(result.confidence * 100).toFixed(1)}%`,
    );
  }

  const sortedResults = results.sort((a, b) => b.confidence - a.confidence);
  console.log("top text:", sortedResults[0]?.text);

  ocr.release();
  capturer.stop();
}
main();

---

rotation

• Readonly rotation: number

Clockwise rotation in degrees ([0, 360)). Meaningful only when OCRDetectionOptions.detectRotation is true; otherwise often 0 or stale.

Notes

• With detectRotation: false (default), ignore this field for geometry.
• 0° = upright reading direction; 180° = upside-down text.

Example

"nodejs";
const { createOCR } = require("ocr");
const { requestScreenCapture } = require("media_projection");

async function main() {
  const ocr = await createOCR();
  const capturer = await requestScreenCapture();
  const img = await capturer.nextImage();

  const results = await ocr.detect(img, {
    detectRotation: true,
  });

  for (const result of results) {
    if (result.rotation !== 0) {
      console.log(`text: ${result.text}`);
      console.log(`rotation: ${result.rotation} deg`);
      console.log(`rotationConfidence: ${result.rotationConfidence}`);
    }
  }

  ocr.release();
  capturer.stop();
}
main();

---

rotationConfidence

• Readonly rotationConfidence: number

Confidence for rotation, 0…1, meaningful only with detectRotation: true.

Notes

• With detectRotation: false, values may be 0 or meaningless.
• Closer to 1 means the angle estimate is trustworthy.

Methods

click

▸ click(): Promise<boolean>

Taps (bounds.centerX, bounds.centerY) on the live device using the accessibility service—equivalent to manually calling click(x, y) at the box center.

Notes

• Requires an enabled accessibility / automation service.
• If the on-screen UI has changed since the screenshot, the tap can miss.
• Prefer capture → detect → click in quick succession.

Example

"nodejs";
const { createOCR } = require("ocr");
const { requestScreenCapture } = require("media_projection");
const { delay } = require("lang");

async function main() {
  const ocr = await createOCR();
  const capturer = await requestScreenCapture();

  await capturer.awaitForImageAvailable();

  const img = capturer.latestImage();
  const results = await ocr.detect(img);

  for (const result of results) {
    if (result.text.includes("OK") || result.text.includes("Confirm")) {
      console.log(`button: ${result.text}`);
      console.log(`center: (${result.bounds.centerX}, ${result.bounds.centerY})`);

      const success = await result.click();
      console.log(`click: ${success ? "ok" : "failed"}`);

      if (success) {
        await delay(1000);
      }
      break;
    }
  }

  ocr.release();
  capturer.stop();
}
main();

More patterns

"nodejs";
const { createOCR } = require("ocr");
const { requestScreenCapture } = require("media_projection");
const { delay } = require("lang");

async function main() {
  const ocr = await createOCR();
  const capturer = await requestScreenCapture();

  const img = await capturer.nextImage();
  const results = await ocr.detect(img);

  const targetText = "Sign in";
  const targetResult = results.find((r) => r.text === targetText);
  if (targetResult && targetResult.confidence > 0.8) {
    await targetResult.click();
  }

  const highConfidenceResults = results.filter((r) => r.confidence > 0.9);
  for (const result of highConfidenceResults) {
    if (result.text.match(/^\d+$/)) {
      await result.click();
      await delay(500);
    }
  }

  const sortedResults = results.sort((a, b) => {
    if (a.bounds.top !== b.bounds.top) {
      return a.bounds.top - b.bounds.top;
    }
    return a.bounds.left - b.bounds.left;
  });

  for (const result of sortedResults) {
    console.log(`tap: ${result.text}`);
    await result.click();
    await delay(300);
  }

  ocr.release();
  capturer.stop();
}
main();

Returns

Promise<boolean>

Resolves to true if the accessibility tap succeeded, false otherwise.