wangyukun/PRA-only-demo
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix: OCR 识别开启 blocks 输出并从块树提取词/行，修复坐标与表格为空

Browse Source
This commit is contained in:
yu 2026-10-29 13:51:51 +08:00
parent 6f008ed8c0
commit 06886c5b31
1 changed files with 40 additions and 19 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

59
src/main/capabilities/ocr-caps.ts
View File

@ -3,21 +3,47 @@ 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 }
function requireImage(args: Record<string, unknown>, nodeId: string): string {
for (const key of ["image_path", "imagePath", "file_path", "filePath", "path", "image"]) {
for (const key of ["image_path", "imagePath", "file_path", "filePath", "path", "image", "pictureSource"]) {
const v = String(args[key] ?? "").trim();
if (v) return v;
}
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 recognize(imagePath: string, nodeId: string) {
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 {
return await worker.recognize(imagePath);
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();
}
@ -25,27 +51,22 @@ async function recognize(imagePath: string, nodeId: string) {
register("OcrOffline_GetOcrTextNew", async (ctx, node, args) => {
const imagePath = requireImage(args, node.id);
const result = await recognize(imagePath, node.id);
const data = result.data as Tesseract.Page & { words?: Array<{ text: string; bbox: { x0: number; y0: number; x1: number; y1: number }; confidence: number }>; lines?: Array<{ words: Array<{ text: string }> }> };
const text = data.text.trim();
ctx.log("OCR 识别完成：" + text.length + " 字符", "info", node.id);
return text;
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;
});
register("OcrOffline_GetOcrTextPosition", async (ctx, node, args) => {
const imagePath = requireImage(args, node.id);
const result = await recognize(imagePath, node.id);
const data = result.data as Tesseract.Page & { words?: Array<{ text: string; bbox: { x0: number; y0: number; x1: number; y1: number }; confidence: number }>; lines?: Array<{ words: Array<{ text: string }> }> };
const words = (data.words ?? []).map(w => ({ 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 }));
ctx.log("OCR 定位完成：" + words.length + " 个词", "info", node.id);
return words;
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;
});
register("OcrOffline_GetOcrTable", async (ctx, node, args) => {
const imagePath = requireImage(args, node.id);
const result = await recognize(imagePath, node.id);
const data = result.data as Tesseract.Page & { words?: Array<{ text: string; bbox: { x0: number; y0: number; x1: number; y1: number }; confidence: number }>; lines?: Array<{ words: Array<{ text: string }> }> };
const lines = (data.lines ?? []).map((line: { words: Array<{ text: string }> }) => line.words.map((w: { text: string }) => w.text).join("\t"));
ctx.log("OCR 表格识别完成：" + lines.length + " 行", "info", node.id);
return lines;
});
const result = await recognizeWithLayout(imagePath, node.id);
const rows = result.textLines.map(line => line.split(/ {2,}/).filter(Boolean).join("\t"));
ctx.log("OCR 表格识别完成：" + rows.length + " 行", "info", node.id);
return rows;
});