@spotterjs/plugin-ocr

中文文档

OCR plugin for spotterjs. It runs PP-OCR-style detection and recognition with ONNX Runtime in Node.js and returns text-line boxes.

Install

npm install @spotterjs/plugin-ocr @spotterjs/core

Quick Start

import { screen } from "@spotterjs/core";
import { createOcr } from "@spotterjs/plugin-ocr";

const cap = screen.capture();
const ocr = await createOcr();
const lines = await ocr.read(cap);

createOcr() creates the OCR client and downloads the default ONNX models on first use unless local models are provided.

Main API

• createOcr(options?): create an OCR client.
• ensureOcrModels(options?): download and verify model files.
• ocr.read(image, options?): return all recognized text lines.
• ocr.findText(image, text, options?): return the first matching line or throw.
• ocr.findAllText(image, text, options?): return all matching lines.

image can be a CaptureImage, encoded image Buffer, or image file path. Read results include text, score, region, box, and center. For read results, score is OCR recognition confidence. Text lookup results also include matchScore, matchAlgorithm, matchKind, query, and matched. score remains OCR recognition confidence; matchScore is the normalized text-query match score.

GPU acceleration uses the same OCR API by passing ONNX Runtime execution providers:

await createOcr({
  executionProviders: ["dml", "cpu"],
});

For small or low-contrast text, enable the lightweight preprocessing pipeline:

await createOcr({
  preprocess: true,
});

Model cache, download sources, local models, private distribution, coordinates, and integration tests are documented in the OCR guide.

License

Learning and non-commercial use are free. Commercial use requires authorization. See LICENSE.