How to Extract Text from Images and Screenshots in Bulk

Copying text out of one screenshot is easy. Doing it for a few thousand support screenshots, receipt photos or saved social posts is not, and the results vary a lot depending on whether the image is a clean UI capture or a skewed phone photo.

Generic OCR settings are tuned for full pages of text, so they often miss the scattered labels in a screenshot or merge columns in a receipt. The Image to Text OCR actor runs Tesseract 5 with a choice of page segmentation modes, returns a confidence score per image, and can return every word with its pixel box.

This guide picks settings for different image types, filters out unreliable results in Python, and uses word boxes to find and redact sensitive text.

Step by step

1. Get public links to the images

   The actor reads images by URL: PNG, JPEG, TIFF, WebP, GIF, BMP and JPEG 2000. HEIC and AVIF photos from phones and SVG are not supported, so convert photos to JPEG first. Files can be up to 100 MB each.

2. Pick a segmentation mode for the image type

   pageSegmentation controls how Tesseract looks for text. auto suits photos of documents and scanned pages. sparse looks for text anywhere and is the best start for UI screenshots, forms and infographics. single-block suits receipts and short snippets, and single-column suits uniform text. If you are unsure, run a small sample of images in two modes and compare confidence scores.

   {
     "urls": [
       "https://github.com/tesseract-ocr/tessdoc/raw/main/images/eurotext.png"
     ],
     "languages": ["eng"],
     "pageSegmentation": "sparse",
     "minWordConfidence": 50
   }

3. Set languages and noise filtering

   Choose up to 4 of the 29 languages in languages; screenshots from a localised app may need two. minWordConfidence drops words Tesseract is unsure about, which helps with compression artefacts, icons mistaken for letters and background patterns. Start around 40-60 and lower it if real words go missing.

4. Keep reliable results and queue the rest

   Each image is one row with text, wordCount and confidence. Very low confidence or very few words usually means a blurry photo, handwriting or an image with no real text. Split results into accepted and needs-review lists.

   import os
   import requests
   
   url = "https://api.apify.com/v2/acts/actor-forge~image-to-text-ocr/run-sync-get-dataset-items"
   payload = {
       "urls": [
           "https://github.com/tesseract-ocr/tessdoc/raw/main/images/eurotext.png",
           "https://raw.githubusercontent.com/tesseract-ocr/test/main/testing/phototest.tif",
       ],
       "pageSegmentation": "sparse",
   }
   resp = requests.post(url, params={"token": os.environ["APIFY_TOKEN"]}, json=payload, timeout=300)
   resp.raise_for_status()
   
   accepted, review = [], []
   for row in resp.json():
       conf = row.get("confidence") or 0
       if row.get("error") or conf < 75 or (row.get("wordCount") or 0) < 3:
           review.append(row)
       else:
           accepted.append(row)
   
   print(len(accepted), "accepted;", len(review), "need review")

5. Find and redact sensitive text with word boxes

   Turn on includeWords to get every recognised word with its confidence and pixel box (left, top, width, height). The exact field names inside each word entry can be checked in your first run's dataset; the snippet below reads them defensively. With Pillow you can then draw filled rectangles over matching words, for example email addresses, before sharing the screenshot.

   import io, re
   import requests
   from PIL import Image, ImageDraw
   
   EMAIL = re.compile(r"[\w.+-]+@[\w-]+\.[\w.]+")
   
   def redact(row, out_path):
       img = Image.open(io.BytesIO(requests.get(row["url"], timeout=60).content)).convert("RGB")
       draw = ImageDraw.Draw(img)
       for w in row.get("words") or []:
           text = w.get("text", "")
           if EMAIL.search(text):
               box = w.get("bbox") or w
               l, t = box["left"], box["top"]
               draw.rectangle([l, t, l + box["width"], t + box["height"]], fill="black")
       img.save(out_path)

6. Mix in scanned PDFs if you need to

   Scanned PDFs work in this actor too, page by page. If your batches regularly mix PDFs and images, OCR to Text uses the same engine and pricing and is set up for mixed document intake.

Sample input

{
  "urls": [
    "https://github.com/tesseract-ocr/tessdoc/raw/main/images/eurotext.png",
    "https://raw.githubusercontent.com/tesseract-ocr/test/main/testing/phototest.tif"
  ],
  "languages": [
    "eng"
  ],
  "pageSegmentation": "sparse",
  "includeWords": true,
  "minWordConfidence": 50
}

Sample output

Each image becomes one row with its file type, word count, confidence and the extracted text, as in the sample below.

Run it from code

cURL

curl -X POST "https://api.apify.com/v2/acts/actor-forge~image-to-text-ocr/run-sync-get-dataset-items" \
  -H "Authorization: Bearer $APIFY_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"urls": ["https://github.com/tesseract-ocr/tessdoc/raw/main/images/eurotext.png", "https://raw.githubusercontent.com/tesseract-ocr/test/main/testing/phototest.tif"], "languages": ["eng"], "pageSegmentation": "sparse", "includeWords": true, "minWordConfidence": 50}'

JavaScript (Node.js)

// Node.js 18+ (built-in fetch). Set APIFY_TOKEN in your environment.
const input = {
  "urls": [
    "https://github.com/tesseract-ocr/tessdoc/raw/main/images/eurotext.png",
    "https://raw.githubusercontent.com/tesseract-ocr/test/main/testing/phototest.tif"
  ],
  "languages": [
    "eng"
  ],
  "pageSegmentation": "sparse",
  "includeWords": true,
  "minWordConfidence": 50
};

const res = await fetch(
  "https://api.apify.com/v2/acts/actor-forge~image-to-text-ocr/run-sync-get-dataset-items",
  {
    method: "POST",
    headers: {
      Authorization: `Bearer ${process.env.APIFY_TOKEN}`,
      "Content-Type": "application/json",
    },
    body: JSON.stringify(input),
  }
);
const items = await res.json();
console.log(items.length, items[0]);

Python

# pip install apify-client   (set APIFY_TOKEN in your environment)
import os
from apify_client import ApifyClient

client = ApifyClient(os.environ["APIFY_TOKEN"])
run_input = {
    "urls": [
        "https://github.com/tesseract-ocr/tessdoc/raw/main/images/eurotext.png",
        "https://raw.githubusercontent.com/tesseract-ocr/test/main/testing/phototest.tif"
    ],
    "languages": [
        "eng"
    ],
    "pageSegmentation": "sparse",
    "includeWords": True,
    "minWordConfidence": 50
}

run = client.actor("actor-forge/image-to-text-ocr").call(run_input=run_input)
for item in client.dataset(run["defaultDatasetId"]).iterate_items():
    print(item)

The run-sync-get-dataset-items endpoint waits for the run (up to 5 minutes) and returns the dataset rows directly. For longer runs, start the run asynchronously and read the dataset when it finishes, as the Python example does. Replace APIFY_TOKEN with your own token from the Apify console.

What it costs

Extracting text from 2,500 support screenshots costs 2,500 x $0.003 = $7.50, with no compute charge on top. Images that fail to download or turn out blank are not charged.

Tips and pitfalls

• Low-resolution thumbnails OCR badly; link to the full-size image whenever you can.
• Images rotated by 90 or 180 degrees are not auto-rotated; slightly skewed photos are fine.
• Handwriting and decorative fonts come out poorly. Use the confidence score rather than assuming every result is usable.
• Charts and pictures with no text return little or nothing; that is expected, not an error.

FAQ

Actors used in this guide

Related guides

All guides