feat: OCR 后端切换飞桨——RapidOCR ONNX（@gutenye/ocr-node，PaddleOCR v4 模型全离线）

src/main/capabilities/ocr-caps.ts

@@ -1,9 +1,18 @@
-import Tesseract from "tesseract.js";
-import fs from "node:fs";
 import { register } from "../engine/registry";
 import { RuntimeError } from "../engine/types";
 
 interface OcrWord { text: string; x: number; y: number; width: number; height: number; confidence: number }
+interface RapidOcrLine { text: string; mean: number; box?: number[][] }
+
+let ocrInstance: unknown = null;
+
+async function getOcr(): Promise<{ detect: (image: string) => Promise<{ texts: RapidOcrLine[] }> }> {
+  if (ocrInstance) return ocrInstance as { detect: (image: string) => Promise<{ texts: RapidOcrLine[] }> };
+  const mod = await import("@gutenye/ocr-node");
+  const instance = await mod.default.create();
+  ocrInstance = instance;
+  return instance;
+}
 
 function requireImage(args: Record<string, unknown>, nodeId: string): string {
   for (const key of ["image_path", "imagePath", "file_path", "filePath", "path", "image", "pictureSource"]) {
@@ -13,60 +22,48 @@ function requireImage(args: Record<string, unknown>, nodeId: string): string {
   throw new RuntimeError("validation_error", "图片路径为空", { nodeId, componentId: nodeId, recommendation: "填写本地图片路径（png/jpg/bmp）" });
 }
 
-async function recognizeWithLayout(imagePath: string, nodeId: string): Promise<{ text: string; words: OcrWord[]; textLines: string[] }> {
-  if (!fs.existsSync(imagePath)) {
-    throw new RuntimeError("file_not_found", "图片不存在：" + imagePath, { nodeId, componentId: nodeId });
-  }
-  const worker = await Tesseract.createWorker("chi_sim+eng");
-  try {
-    const result = await worker.recognize(imagePath, {}, { text: true, blocks: true });
-    const data = result.data as {
-      text?: string;
-      blocks?: Array<{
-        paragraphs?: Array<{
-          lines?: Array<{
-            words?: Array<{ text: string; confidence: number; bbox: { x0: number; y0: number; x1: number; y1: number } }>;
-          }>;
-        }>;
-      }>;
-    };
-    const words: OcrWord[] = [];
-    const textLines: string[] = [];
-    for (const block of data.blocks ?? []) {
-      for (const para of block.paragraphs ?? []) {
-        for (const line of para.lines ?? []) {
-          const lineWords = line.words ?? [];
-          textLines.push(lineWords.map(w => w.text).join(" "));
-          for (const w of lineWords) {
-            words.push({ text: w.text, x: w.bbox.x0, y: w.bbox.y0, width: w.bbox.x1 - w.bbox.x0, height: w.bbox.y1 - w.bbox.y0, confidence: w.confidence });
-          }
-        }
-      }
-    }
-    return { text: (data.text ?? "").trim(), words, textLines };
-  } finally {
-    await worker.terminate();
-  }
+function boxToRect(box: number[][] | undefined): { x: number; y: number; width: number; height: number } {
+  const points = box ?? [];
+  const xs = points.map(p => p[0]);
+  const ys = points.map(p => p[1]);
+  const minX = Math.min(...(xs.length ? xs : [0]));
+  const minY = Math.min(...(ys.length ? ys : [0]));
+  const maxX = Math.max(...(xs.length ? xs : [0]));
+  const maxY = Math.max(...(ys.length ? ys : [0]));
+  return { x: minX, y: minY, width: maxX - minX, height: maxY - minY };
 }
 
 register("OcrOffline_GetOcrTextNew", async (ctx, node, args) => {
   const imagePath = requireImage(args, node.id);
-  const result = await recognizeWithLayout(imagePath, node.id);
-  ctx.log("OCR 识别完成：" + result.text.length + " 字符 / " + result.words.length + " 词", "info", node.id);
-  return result.text;
+  const ocr = await getOcr();
+  const result = await ocr.detect(imagePath);
+  const text = result.texts.map(t => t.text).join("\n");
+  ctx.log("OCR 识别完成：" + text.length + " 字符 / " + result.texts.length + " 行", "info", node.id);
+  return text;
 });
 
 register("OcrOffline_GetOcrTextPosition", async (ctx, node, args) => {
   const imagePath = requireImage(args, node.id);
-  const result = await recognizeWithLayout(imagePath, node.id);
-  ctx.log("OCR 定位完成：" + result.words.length + " 个词", "info", node.id);
-  return result.words;
+  const ocr = await getOcr();
+  const result = await ocr.detect(imagePath);
+  const words = result.texts.map(t => {
+    const rect = boxToRect(t.box);
+    return { text: t.text, x: rect.x, y: rect.y, width: rect.width, height: rect.height, confidence: t.mean };
+  });
+  ctx.log("OCR 定位完成：" + words.length + " 行", "info", node.id);
+  return words;
 });
 
 register("OcrOffline_GetOcrTable", async (ctx, node, args) => {
   const imagePath = requireImage(args, node.id);
-  const result = await recognizeWithLayout(imagePath, node.id);
-  const rows = result.textLines.map(line => line.split(/ {2,}/).filter(Boolean).join("\t"));
+  const ocr = await getOcr();
+  const result = await ocr.detect(imagePath);
+  const sorted = [...result.texts].sort((a, b) => {
+    const ay = a.box ? Math.min(...a.box.map(p => p[1])) : 0;
+    const by = b.box ? Math.min(...b.box.map(p => p[1])) : 0;
+    return ay - by;
+  });
+  const rows = sorted.map(t => t.text);
   ctx.log("OCR 表格识别完成：" + rows.length + " 行", "info", node.id);
   return rows;
 });