Complete PDF Guide

Practical Snippet

Key points: (1) distinguish electronic from scanned documents — a scan text layer is OCR-guessed at 85–98% accuracy so search misses and copied identifiers are corrupted, and any content editing needs OCR with manual spot-checks of amounts and identifiers first; (2) compressing a PDF means recompressing images (300dpi to 150dpi, PNG to lossy JPEG, saving 60–80%) rather than the document structure, since images are about 95% of a scan; (3) encryption comes as an open password or a permissions password, and cannot-open is a different problem from opens-but-cannot-copy; (4) merging and splitting break bookmarks and internal links, so verify every bookmark page after merging — out-of-range numbers mean the mapping is wrong.

// ① 判断 PDF 类型：是扫描件还是电子件
// 扫描件几乎没有可提取的文字对象
async function pdfKind(file) {
  const buf = new Uint8Array(await file.arrayBuffer())
  const text = new TextDecoder('latin1').decode(buf)
  const hasFont = //Font/.test(text)
  const hasImage = //Subtypes*/Image/.test(text)
  const pageCount = (text.match(//Types*/Page[^s]/g) || []).length
  // 有 Image 但无 Font ⇒ 大概率是扫描件（图片 + 极少的 OCR 文字层）
  return { pageCount, hasFont, hasImage, likelyScanned: hasImage && !hasFont }
}

// ② 压缩：压图片而不是压文档结构
// ✅ 有效：300dpi → 150dpi，PNG → 有损 JPEG（体积降 60–80%）
// ❌ 无效：只做 "compress document"（字体与结构本来就只占几十 KB）
const COMPRESS_OPTIONS = {
  imageDpi: 150,        // 300 → 150，文字型 PDF 可再降到 100
  jpegQuality: 0.75,    // 0.6–0.8 之间肉眼几乎无差
  removeMetadata: true, // 元数据通常只占几 KB
  // ❌ 不要启用 "optimize for web" 之类会降文字清晰度的选项
}

// ③ 加密分两种，问题不同
// 打开密码：不给就打不开
function encryptWithOpenPassword(doc, pwd) {
  return { userPassword: pwd, ownerPassword: pwd, allowPrint: true, allowCopy: false }
}
// 权限密码：不给也能打开，但限制操作
function encryptWithPermissions(doc) {
  return { userPassword: '', ownerPassword: '', allowPrint: false, allowCopy: false }
}
// 排查「打不开」还是「不能复制」，先分清是哪一种

// ④ 合并：结构化操作 + 合并后校验书签
async function mergeSafely(files) {
  const merged = await mergePreservingOutline(files)
  // 🔴 书签可能指向错误的页码 ⇒ 合并后必须逐个检查
  const outline = await readOutline(merged)
  const pageCount = await countPages(merged)
  const broken = outline.filter(item => item.page > pageCount)
  if (broken.length) {
    console.warn('书签越界：', broken)    // 越界说明页码映射错了
  }
  return { doc: merged, brokenBookmarks: broken.length }
}

// ⑤ 扫描件：OCR 后抽查关键字段再批量处理
const CRITICAL_FIELDS = ['amount', 'invoice_no', 'date']   // 金额/编号/日期
function shouldSpotCheck(text) {
  return CRITICAL_FIELDS.some(f => text.includes(f))
}
// OCR 准确率通常 85–98%，错误集中在低频字符、标点与数字上