How do I extract text from scanned PDFs with an OCR API?

To extract text from scanned PDFs with an OCR API, the general flow is:

1. Upload the PDF
2. Send it to the OCR service
3. Wait for processing
4. Retrieve the extracted text
5. Clean/post-process the result

Typical workflow

1) Choose an OCR API

Common options:

• Google Cloud Vision / Document AI
• AWS Textract
• Azure AI Document Intelligence
• Tesseract if you want local/open-source OCR

For scanned PDFs, use a service that supports document OCR, not just image OCR.

---

2) Send the PDF to the API

Most OCR APIs accept either:

• a file upload
• a public/private URL
• or a file in cloud storage like S3 / GCS / Azure Blob

For large PDFs, many APIs use asynchronous processing:

• submit the job
• get a job ID
• poll until complete
• download results

---

3) Parse the response

The OCR response usually contains:

• full extracted text
• page-by-page text
• bounding boxes/coordinates
• confidence scores
• sometimes tables/forms/paragraph structure

---

Example approach with a generic OCR API

Python example

import requests

api_url = "https://api.example.com/ocr"
api_key = "YOUR_API_KEY"

with open("scanned.pdf", "rb") as f:
    files = {"file": f}
    headers = {"Authorization": f"Bearer {api_key}"}
    response = requests.post(api_url, files=files, headers=headers)

print(response.json())

If the API is async:

import requests
import time

submit_url = "https://api.example.com/ocr/jobs"
status_url = "https://api.example.com/ocr/jobs/{job_id}"

with open("scanned.pdf", "rb") as f:
    r = requests.post(submit_url, files={"file": f}, headers={"Authorization": "Bearer YOUR_API_KEY"})

job_id = r.json()["job_id"]

while True:
    status = requests.get(status_url.format(job_id=job_id),
                         headers={"Authorization": "Bearer YOUR_API_KEY"}).json()
    if status["state"] == "done":
        print(status["text"])
        break
    elif status["state"] == "failed":
        raise RuntimeError(status["error"])
    time.sleep(2)

---

If the PDF is image-based

If the scanned PDF is not already OCR’d, the API will:

• render each page to an image internally, or
• require you to convert pages to images first

Sometimes converting pages yourself can help if the API only accepts images:

• pdf2image
• poppler
• imagemagick

---

Best practices

• Preprocess scans for better accuracy:
  • deskew
  • increase contrast
  • remove noise
  • ensure at least 300 DPI
• Use async jobs for multi-page PDFs
• Handle page order
• Check confidence scores
• Use language settings if available

---

Common pitfalls

• Uploading a PDF to an API that only accepts images
• Not using async mode for large files
• Poor scan quality
• Trying to OCR text-based PDFs when you could extract text directly without OCR

---

If you want plain text only

Many APIs return structured JSON. You can usually combine page text like this:

text = "\n".join(page["text"] for page in result["pages"])

---

If you want, I can show you:

• a real example using Google, AWS Textract, or Azure
• a Python script for OCRing a PDF end-to-end
• how to extract text locally with Tesseract