10 Tips to Improve OCR Accuracy for Better PDF Text Recognition

Even the best OCR software produces poor results when given a bad input image. The most important factor in OCR accuracy isn’t which software you choose — it’s the quality of your scanned document.

Here are ten proven techniques to maximize OCR accuracy.

---

1. Scan at the Right Resolution

Resolution is the single most important factor. Always scan at:

• Minimum: 200 DPI (barely acceptable for clean documents)
• Recommended: 300 DPI (good balance of quality and file size)
• Best: 600 DPI (for faded, small-font, or degraded documents)

Higher DPI means more pixels per character, giving the OCR engine more information to work with.

Most modern smartphone cameras exceed 300 DPI equivalent when held 30–40cm from a document.

---

2. Use Grayscale or Black & White Scans

For text documents, color scanning adds noise without benefit. The OCR engine only needs to distinguish dark text from light background.

• Grayscale (8-bit): Best for most documents — preserves subtle contrast
• Black & White (1-bit): Fastest processing; best for high-contrast printed text
• Color: Use only when the document has meaningful color content (charts, forms with colored fields)

---

3. Ensure Proper Lighting (For Camera Scans)

When photographing documents with a phone:

• Use bright, even lighting — stand near a window during the day
• Avoid flash — it creates hot spots and glare
• No shadows — don’t hold the phone over the document; use a stand or lean it against a wall
• Diffuse light is best — overcast sky or multiple light sources

The Google PhotoScan app uses multiple shots to eliminate glare automatically.

---

4. Deskew and Straighten Documents

Even a 1–2° tilt significantly reduces OCR accuracy. Most OCR tools have automatic deskew, but you can also:

• Use a document feeder on a flatbed scanner (keeps pages perfectly flat)
• Use OCRmyPDF’s --deskew option for automatic correction
• In Photoshop/GIMP, use the Straighten tool before OCR

---

5. Increase Contrast Before OCR

Low contrast between text and background is the number one cause of OCR errors on old or faded documents. Fix this with image preprocessing:

from PIL import Image, ImageEnhance

img = Image.open('faded_document.jpg').convert('L')  # grayscale
enhancer = ImageEnhance.Contrast(img)
img_high_contrast = enhancer.enhance(2.5)  # increase contrast by 2.5x
img_high_contrast.save('enhanced.jpg')

Or use free tools like GIMP: Colors → Levels → drag the black/white input points inward.

---

6. Remove Background Noise

Spotted or textured backgrounds (recycled paper, watermarked letterheads) confuse OCR. Preprocessing:

import cv2
import numpy as np

img = cv2.imread('noisy.jpg', cv2.IMREAD_GRAYSCALE)
# Threshold to binary (Otsu's method - automatic threshold selection)
_, binary = cv2.threshold(img, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)
cv2.imwrite('cleaned.jpg', binary)

OCRmyPDF’s --clean flag uses unpaper to remove background artifacts automatically.

---

7. Select the Correct Language

Always specify the document language. An OCR engine using an English dictionary to read French will produce errors on accented characters.

# Tesseract: specify language
tesseract document.pdf output -l fra

# Multi-language document
tesseract document.pdf output -l fra+eng

Most online OCR tools have a language dropdown — don’t leave it on “Auto” if you know the language.

---

8. Use the Right OCR Mode for Your Document

Tesseract has multiple “page segmentation modes” (PSM):

# Default (auto segment)
tesseract doc.png out -l eng --psm 3

# Single column of text
tesseract doc.png out --psm 4

# Single line (good for forms)
tesseract doc.png out --psm 7

# Single word (good for labels)
tesseract doc.png out --psm 8

For most documents, the default (auto) works well. For forms, labels, and single-column text, specifying the PSM improves accuracy.

---

9. Use Post-Processing Spell Check

OCR output always has some errors. Running a spell checker on the output catches many of them:

from spellchecker import SpellChecker

spell = SpellChecker()
words = ocr_text.split()
misspelled = spell.unknown(words)
for word in misspelled:
    correction = spell.correction(word)
    if correction:
        ocr_text = ocr_text.replace(word, correction)

For domain-specific documents (legal, medical), add a custom dictionary.

---

10. Split Large Documents

For multi-page PDFs, processing page by page gives better results than processing the whole document at once:

# Split with pdftk
pdftk input.pdf burst output page_%04d.pdf

# Then OCR each page
for file in page_*.pdf; do
  ocrmypdf --force-ocr "$file" "ocr_${file}"
done

---

Quick Reference Checklist

• 300 DPI scan resolution
• Grayscale mode for text documents
• Document flat and straight
• Even lighting, no shadows
• Correct language selected
• Background noise removed if present
• Contrast enhanced for faded documents

Follow these tips and you’ll see OCR accuracy jump from 85–90% to 97–99% on most standard documents.