test-ocr.mjs

// 头less 验证：用扩展里 vendor 的同一套 Tesseract 资源 + 同样的白名单/PSM 配置，
// 对合成的比赛计时器图做 OCR，确认"裁出的 ROI → OCR 数字"这一半链路可行且准确。
// （tabCapture 截帧那一半必须在真实 Chrome 里手动测。）

import { createCanvas } from "@napi-rs/canvas";
import { createWorker } from "tesseract.js";
import { fileURLToPath } from "node:url";
import path from "node:path";

const __dirname = path.dirname(fileURLToPath(import.meta.url));
const vendor = path.join(__dirname, "extension", "vendor");

function makeClockImage(text, scale = 3) {
  // 模拟扩展里裁出的小 ROI 再放大 ROI_SCALE 倍的效果
  const baseW = 120, baseH = 48;
  const c = createCanvas(baseW * scale, baseH * scale);
  const ctx = c.getContext("2d");
  ctx.fillStyle = "#0b1a2b";              // 深色比分条背景
  ctx.fillRect(0, 0, c.width, c.height);
  ctx.fillStyle = "#ffffff";
  ctx.font = `bold ${30 * scale}px Arial`;
  ctx.textBaseline = "middle";
  ctx.fillText(text, 8 * scale, c.height / 2);
  return c;
}

async function run() {
  const cases = ["67:23", "45:00", "90:12", "12:08"];
  // Node 下用 tesseract.js 自带的 node worker/core；langPath 指向扩展 vendor 的语言包，
  // 以验证我们打包进扩展的 eng.traineddata.gz 是有效的。
  const worker = await createWorker("eng", 1, {
    langPath: vendor,
    gzip: true,
  });
  await worker.setParameters({
    tessedit_char_whitelist: "0123456789:+",
    tessedit_pageseg_mode: "7",
  });

  let pass = 0;
  for (const expected of cases) {
    const canvas = makeClockImage(expected);
    const png = canvas.toBuffer("image/png");
    const t0 = performance.now();
    const { data } = await worker.recognize(png);
    const ms = Math.round(performance.now() - t0);
    const got = (data.text || "").replace(/\s+/g, "");
    const ok = got === expected;
    if (ok) pass++;
    console.log(`${ok ? "✅" : "❌"} 期望 "${expected}" → OCR "${got}"  (${ms}ms, conf ${Math.round(data.confidence)})`);
  }
  await worker.terminate();
  console.log(`\n结果：${pass}/${cases.length} 通过`);
  process.exit(pass === cases.length ? 0 : 1);
}

run().catch((e) => { console.error(e); process.exit(2); });