Image OCR Expert

Expert in extracting, processing, and structuring text from images using OCR tools and techniques.

Description

This skill provides specialized knowledge for extracting text from images, including:

• Tool and library selection by use case (Tesseract, EasyOCR, PaddleOCR, cloud APIs)
• Image preprocessing to maximize OCR accuracy
• Post-processing and structuring of extracted text
• Handling handwriting, receipts, invoices, documents, screenshots
• Multilingual OCR and special character support
• Integration into Python/Node.js/cloud pipelines

Triggers: ocr, extract text from image, image to text, read text image, optical character recognition, tesseract, easyocr, paddleocr, textract, vision api, document extraction, screenshot text, invoice ocr, receipt ocr, handwriting recognition, image text extraction

Tool Selection Guide

Decision Tree

Is accuracy critical and budget available?
├─ YES → Google Vision API or AWS Textract
└─ NO → Local solution
    ├─ CJK (Chinese/Japanese/Korean) or tables? → PaddleOCR
    ├─ General photos or multiple languages? → EasyOCR
    ├─ Simple printed English docs? → Tesseract
    └─ PDF documents with structure? → Docling or Surya

Python Implementations

Tesseract (pytesseract)

import pytesseract
from PIL import Image
import cv2
import numpy as np

def extract_text_tesseract(image_path: str, lang: str = "eng") -> str:
    """Extract text using Tesseract. Best for clean printed documents."""
    image = Image.open(image_path)

    # Config: --psm 6 = assume uniform block of text
    config = "--psm 6 --oem 3"
    text = pytesseract.image_to_string(image, lang=lang, config=config)
    return text.strip()

def extract_with_confidence(image_path: str) -> list[dict]:
    """Extract text with bounding boxes and confidence scores."""
    image = Image.open(image_path)
    data = pytesseract.image_to_data(image, output_type=pytesseract.Output.DICT)

    results = []
    for i, word in enumerate(data["text"]):
        if word.strip() and int(data["conf"][i]) > 30:
            results.append({
                "text": word,
                "confidence": data["conf"][i],
                "bbox": {
                    "x": data["left"][i],
                    "y": data["top"][i],
                    "width": data["width"][i],
                    "height": data["height"][i],
                }
            })
    return results

# Install: pip install pytesseract pillow
# System: apt install tesseract-ocr (Linux) / brew install tesseract (Mac)

EasyOCR

import easyocr
from pathlib import Path

def extract_text_easyocr(
    image_path: str,
    languages: list[str] = ["en"],
    detail: bool = False
) -> str | list:
    """
    Extract text using EasyOCR. Best for photos and multiple languages.
    languages: ['en'], ['en', 'es'], ['ch_sim', 'en'], etc.
    """
    reader = easyocr.Reader(languages, gpu=False)  # gpu=True if CUDA available
    results = reader.readtext(image_path)

    if not detail:
        # Return plain text sorted by vertical position
        results_sorted = sorted(results, key=lambda x: x[0][0][1])
        return "\n".join([text for _, text, conf in results_sorted if conf > 0.3])

    return [
        {
            "text": text,
            "confidence": round(conf, 3),
            "bbox": bbox,
        }
        for bbox, text, conf in results
    ]

# Install: pip install easyocr

PaddleOCR (best for CJK and tables)

from paddleocr import PaddleOCR
import json

def extract_text_paddle(
    image_path: str,
    lang: str = "en",  # "en", "ch", "japan", "korean", "es", etc.
    use_angle_cls: bool = True,
) -> str:
    """Extract text using PaddleOCR. Best for CJK and structured documents."""
    ocr = PaddleOCR(use_angle_cls=use_angle_cls, lang=lang, show_log=False)
    result = ocr.ocr(image_path, cls=True)

    lines = []
    if result and result[0]:
        # Sort by y position (top to bottom)
        items = sorted(result[0], key=lambda x: x[0][0][1])
        lines = [item[1][0] for item in items if item[1][1] > 0.3]

    return "\n".join(lines)

# Install: pip install paddlepaddle paddleocr

Google Vision API

from google.cloud import vision
import io

def extract_text_google_vision(image_path: str) -> dict:
    """
    Extract text using Google Vision API.
    Requires: GOOGLE_APPLICATION_CREDENTIALS env var set.
    """
    client = vision.ImageAnnotatorClient()

    with io.open(image_path, "rb"