Using pytesseract and Pillow for OCR: Installation, Configuration, and Accuracy Improvement Techniques

This article provides a step‑by‑step tutorial for performing OCR using the Tesseract engine with Python's pytesseract and Pillow libraries.

1. Install required libraries – Install Tesseract OCR (Windows: download installer; macOS: brew install tesseract; Linux: sudo apt-get install tesseract-ocr) and then install the Python packages with pip install pytesseract pillow.

2. Set Tesseract path (Windows only) – Add the executable path in Python:

import pytesseract
pytesseract.pytesseract.tesseract_cmd = r'C:\Program Files\Tesseract-OCR\tesseract.exe'

3. Read image and perform OCR – Example code loads an image and extracts text:

import pytesseract
from PIL import Image
# image_path = 'id_card.jpg'
image = Image.open(image_path)
text = pytesseract.image_to_string(image)  # Simplified Chinese
print(text)

4. Process OCR results – Use regular expressions to extract specific fields such as name and ID number:

import re
name_pattern = re.compile(r'姓名\s*([\u4e00-\u9fa5]+)')
id_pattern = re.compile(r'公民身份号码\s*(\d{18})')
name_match = name_pattern.search(text)
id_match = id_pattern.search(text)
if name_match:
    print(f"姓名: {name_match.group(1)}")
else:
    print("未找到姓名")
if id_match:
    print(f"公民身份号码: {id_match.group(1)}")
else:
    print("未找到公民身份号码")

5. Improve OCR accuracy – Techniques include image preprocessing (brightness, contrast, denoising, sharpening), using appropriate language models, and block‑wise recognition. Example preprocessing code:

from PIL import ImageEnhance
image = Image.open(image_path)
enhancer = ImageEnhance.Brightness(image)
image_enhanced = enhancer.enhance(1.5)
enhancer = ImageEnhance.Contrast(image_enhanced)
image_enhanced = enhancer.enhance(1.5)
text = pytesseract.image_to_string(image_enhanced)
print(text)

Additional methods cover text region detection with OpenCV, selecting high‑quality OCR engines (Tesseract, Google Cloud Vision, ABBYY FineReader), applying language models and dictionaries, block‑wise OCR, and post‑processing such as spell checking and regex extraction.

Summary – By combining proper installation, configuration, preprocessing, region detection, engine selection, and post‑processing, OCR recognition accuracy can be substantially enhanced.