OCRmyPDF 教學：掃描文件轉換為可搜尋 PDF/A 檔案，支援側車文字擷取與批次處理

In this tutorial, we build an advanced, self-contained OCRmyPDF workflow. We start by installing the required system and Python dependencies, then create a synthetic image-only PDF for scanning so we can test OCR without relying on external files. From there, we use OCRmyPDF’s real public API to convert scanned documents into searchable PDFs, generate PDF/A outputs, extract sidecar text, validate the results, compare file sizes, tune Tesseract settings, clean noisy scans, handle already-OCRed files, process images with DPI hints, run OCR in memory, and batch-process multiple PDFs. Through this workflow, we understand how OCRmyPDF can serve as a practical document digitization pipeline for archival, search, extraction, and automated processing tasks. Installing OCRmyPDF System Dependencies Copy CodeCopiedUse a different Browserimport io import os import re import sys import time import shutil import logging import textwrap import subprocess from pathlib import Path INSTALL_JBIG2 = True def sh(cmd: str, check: bool = True) -> int: """Run a shell command, echo it, and show the tail of its output.""" print(f" $ {cmd}") r = subprocess.run(cmd, shell=True, text=True, stdout=subprocess.PIPE, stderr=subprocess.STDOUT) if r.stdout and r.stdout.strip(): for ln in r.stdout.strip().splitlines()[-12:]: print(" " + ln) if check and r.returncode != 0: raise RuntimeError(f"Command failed ({r.returncode}): {cmd}") return r.returncode def install_dependencies() -> None: """Install OCRmyPDF's system + Python dependencies for Colab/Ubuntu.""" apt_pkgs = ( "tesseract-ocr tesseract-ocr-eng tesseract-ocr-osd " "tesseract-ocr-deu tesseract-ocr-fra " "ghostscript unpaper pngquant poppler-utils qpdf" ) sh("apt-get update -qq", check=False) sh(f"DEBIAN_FRONTEND=noninteractive apt-get install -y -qq {apt_pkgs}") sh(f'"{sys.executable}" -m pip install -q --upgrade ocrmypdf img2pdf "pillow<12"') if INSTALL_JBIG2 and shutil.which("jbig2") is None: try: build_pkgs = ("autoconf automake libtool pkg-config " "libleptonica-dev zlib1g-dev build-essential git") sh(f"DEBIAN_FRONTEND=noninteractive apt-get install -y -qq {build_pkgs}") sh("rm -rf /tmp/jbig2enc && " "git clone -q https://github.com/agl/jbig2enc.git /tmp/jbig2enc") sh("cd /tmp/jbig2enc && ./autogen.sh >/dev/null 2>&1 && " "./configure >/dev/null 2>&1 && make -j2 >/dev/null 2>&1 && " "make install >/dev/null 2>&1 && ldconfig") print(" jbig2enc:", "installed" if shutil.which("jbig2") else "built, but binary not on PATH") except Exception as e: print(" jbig2enc build skipped (optional):", e) def ensure_installed() -> None: have_tools = bool(shutil.which("tesseract") and shutil.which("gs")) try: import ocrmypdf import img2pdf from PIL import Image have_py = True except Exception: have_py = False if have_tools and have_py: print("Dependencies already present — skipping installation.") else: print("Installing dependencies (first run can take a few minutes)...") install_dependencies() ensure_installed() We set up the complete OCRmyPDF environment for Google Colab by importing the required standard libraries and defining the installation workflow. We install system tools such as Tesseract, Ghostscript, unpaper, pngquant, poppler, and qpdf, along with Python packages like OCRmyPDF, img2pdf, and Pillow. We also optionally build jbig2enc so that advanced PDF optimization can produce smaller outputs for scanned documents. Loading OCRmyPDF and Building Synthetic Scans Copy CodeCopiedUse a different Browserdef _purge(*prefixes): for name in [m for m in list(sys.modules) if any(m == p or m.startswith(p + ".") for p in prefixes)]: del sys.modules[name] def _load_ocrmypdf(): _purge("PIL", "ocrmypdf") import ocrmypdf return ocrmypdf try: ocrmypdf = _load_ocrmypdf() except ImportError as e: if "_Ink" in str(e) or "PIL" in str(e): print("Repairing an incompatible Pillow (reinstalling pillow<12)...") sh(f'"{sys.executable}" -m pip install -q --force-reinstall "pillow<12"') try: ocrmypdf = _load_ocrmypdf() print("Pillow repaired — continuing without a restart.") except Exception: raise RuntimeError( "Pillow is still incompatible in this session. Use the Colab menu: " "Runtime > Restart session, then run this cell again." ) else: raise from ocrmypdf.exceptions import ( ExitCode, PriorOcrFoundError, EncryptedPdfError, MissingDependencyError, TaggedPDFError, DigitalSignatureError, DpiError, InputFileError, UnsupportedImageFormatError, ) from ocrmypdf.helpers import check_pdf from ocrmypdf.pdfa import file_claims_pdfa import img2pdf from PIL import Image, ImageDraw, ImageFont, ImageFilter logging.basicConfig(level=logging.WARNING, format="%(levelname)s: %(message)s") logging.getLogger("ocrmypdf").setLevel(logging.WARNING) logging.getLogger("pdfminer").setLevel(logging.ERROR) logging.getLogger("PIL").setLevel(logging.WARNING) SAMPLE_TEXT_PAGES = [ "Optical Character Recognition, commonly abbreviated as OCR, is the " "process of converting images of typed or printed text into machine " "encoded text. This page was generated as a synthetic scan so that the " "OCRmyPDF pipeline has something realistic to recognize and search.", "On 14 March 2026 the archive contained 1,482 pages across 37 folders. " "Roughly 92 percent of those pages were scanned at 200 to 300 dots per " "inch. The remaining 8 percent were skewed and required deskewing before " "any reliable recognition was possible.", "After OCRmyPDF finishes, the output is a searchable PDF/A file. You can " "select text, copy it, and run full text search across thousands of " "documents. The original image resolution is preserved while a hidden " "text layer is placed accurately underneath the page image.", ] def _find_font(): for cand in ( "/usr/share/fonts/truetype/dejavu/DejaVuSans.ttf", "/usr/share/fonts/truetype/liberation/LiberationSans-Regular.ttf", ): if os.path.exists(cand): return cand return None _FONT_PATH = _find_font() FONT = ImageFont.truetype(_FONT_PATH, 40) if _FONT_PATH else ImageFont.load_default() def _add_speckle(img, n=6000, dark=60): """Sprinkle light dark specks to imitate scanner noise (motivates --clean).""" import random px = img.load() w, h = img.size for _ in range(n): px[random.randint(0, w - 1), random.randint(0, h - 1)] = random.randint(0, dark) return img def render_page(text, skew=False): """Render one A4 page (1654x2339 px ≈ 200 DPI) of dark text on white.""" W, H = 1654, 2339 img = Image.new("L", (W, H), 255) draw = ImageDraw.Draw(img) draw.multiline_text((150, 180), textwrap.fill(text, width=58), fill=25, font=FONT, spacing=18) if skew: img = img.rotate(6, resample=Image.BICUBIC, expand=False, fillcolor=255) img = img.filter(ImageFilter.GaussianBlur(0.6)) img = _add_speckle(img) return img def build_scanned_pdf(pdf_path: Path, pages_text, skew_index=1): """Render pages to PNGs and wrap them losslessly into an image-only PDF.""" pngs = [] for i, text in enumerate(pages_text): img = render_page(text, skew=(i == skew_index)) p = pdf_path.parent / f"_pg_{pdf_path.stem}_{i}.png" img.save(p, format="PNG", dpi=(200, 200)) pngs.append(str(p)) with open(pdf_path, "wb") as f: f.write(img2pdf.convert(pngs)) for p in pngs: os.remove(p) return pdf_path def do_ocr(input_file, output_file, **kw): """Wrapper around ocrmypdf.ocr() that disables the progress bar and times it.""" kw.setdefault("progress_bar", False) t0 = time.perf_counter() rc = ocrmypdf.ocr(input_file, output_file, **kw) return rc, time.perf_counter() - t0 def tokens(s: str): return re.findall(r"[a-z0-9]+", s.lower()) def kb(path) -> str: return f"{Path(path).stat().st_size / 1024:,.1f} KB" def banner(title: str): line = "─" * 74 print(f"\n{line}\n {title}\n{line}") We safely load OCRmyPDF and repair Pillow compatibility issues if they appear in the Colab runtime. We import OCRmyPDF exceptions, PDF validation helpers, img2pdf, and Pillow utilities used throughout the tutorial. We also define the sample document text and helper functions for rendering synthetic scanned pages, adding scanner-like noise, building ima