anyocr 0.2.0

A lightweight, unified OCR toolkit with a one-liner API. Supports Surya, EasyOCR, PaddleOCR, Tesseract, and Vision LLMs through a single interface.

anyocr

A lightweight, unified OCR toolkit with a one-liner API. Supports multiple backends (Surya, EasyOCR, PaddleOCR, Tesseract) through a single interface.

Runs completely offline after first model download. OCR models are cached locally and no internet connection is required for subsequent use.

Author: Viet-Anh Nguyen | GitHub

---

Features

• One-liner API -- anyocr.read("image.png") and you're done
• Multiple backends -- EasyOCR, PaddleOCR, Tesseract, all through one interface
• Auto-detection -- automatically selects the best available backend
• Flexible input -- accepts file paths, numpy arrays, PIL Images, bytes, or URLs
• Batch processing -- process multiple images efficiently
• Preprocessing -- built-in image enhancement for better accuracy
• Type-safe -- full type hints and dataclass results

Installation

Install the core package (no OCR backend):

pip install anyocr

Then install your preferred backend(s):

# EasyOCR -- recommended for general use
pip install anyocr[easyocr]

# PaddleOCR -- best for CJK languages
pip install anyocr[paddle]

# Tesseract -- widest language support (requires tesseract system package)
pip install anyocr[tesseract]

# Install all backends
pip install anyocr[all]

Tesseract system dependency

If using the Tesseract backend, you also need the system binary:

# Ubuntu/Debian
sudo apt install tesseract-ocr

# macOS
brew install tesseract

# Windows -- download from https://github.com/UB-Mannheim/tesseract/wiki

Quick Start

import anyocr

# Read text from an image (auto-selects best available backend)
result = anyocr.read("document.png")
print(result.text)

# Use a specific backend
result = anyocr.read("document.png", backend="easyocr")

# Multi-language support
result = anyocr.read("document.png", lang=["en", "vi"])

# Access detailed results
for line in result.lines:
    print(f"Text: {line.text}")
    print(f"  BBox: {line.bbox}")
    print(f"  Confidence: {line.confidence:.2f}")

# Batch processing
results = anyocr.read_batch(["page1.png", "page2.png", "page3.png"])

# Check available backends
print(anyocr.available_backends())  # e.g. ['easyocr', 'tesseract']

# Set default backend
anyocr.set_default_backend("easyocr")

Flexible image input

import anyocr
from PIL import Image
import numpy as np

# File path
result = anyocr.read("photo.jpg")

# PIL Image
img = Image.open("photo.jpg")
result = anyocr.read(img)

# Numpy array
arr = np.array(img)
result = anyocr.read(arr)

# Bytes
with open("photo.jpg", "rb") as f:
    result = anyocr.read(f.read())

# URL
result = anyocr.read("https://example.com/image.png")

Preprocessing

import anyocr

# Enable preprocessing for scanned documents
result = anyocr.read(
    "scan.png",
    do_preprocessing=True,
    do_deskew=True,
    do_enhance_contrast=True,
    contrast_factor=1.8,
)

Backend Comparison

Feature	EasyOCR	PaddleOCR	Tesseract
Speed	Medium	Fast	Fast
Accuracy	High	Very High	Good
Languages	80+	80+	100+
GPU support	Yes	Yes	No
CJK quality	Good	Excellent	Fair
Setup	pip only	pip only	System binary required
Model size	Large	Medium	Small
Best for	General use	CJK / documents	Legacy systems

Supported Languages

All backends support English. For other languages:

• EasyOCR: 80+ languages including Vietnamese, Chinese, Japanese, Korean, Thai, Arabic, Hindi, and more. Full list
• PaddleOCR: 80+ languages with excellent CJK support. Full list
• Tesseract: 100+ languages. Install additional language packs via your system package manager. Full list

API Reference

anyocr.read(image, backend=None, lang=None, do_preprocessing=False, **kwargs)

Read text from a single image.

• image -- file path, numpy array, PIL Image, bytes, or URL
• backend -- "easyocr", "paddleocr", "tesseract", or None (auto-select)
• lang -- list of language codes, e.g. ["en", "vi"]
• do_preprocessing -- enable image preprocessing pipeline
• Returns: OCRResult

anyocr.read_batch(images, **kwargs)

Read text from multiple images. Same options as read().

anyocr.available_backends()

Returns a list of installed backend names.

anyocr.set_default_backend(name)

Set the default backend for all subsequent calls.

OCRResult

• .text -- full recognized text (lines joined by newlines)
• .lines -- list of TextLine objects
• .confidence -- average confidence score
• .backend -- name of the backend used
• .language -- language codes used

TextLine

• .text -- recognized text
• .bbox -- bounding box as (x_min, y_min, x_max, y_max)
• .confidence -- confidence score (0.0 to 1.0)

Local-First / Edge AI

This package is designed to work completely offline. After initial model download, no internet connection is required.

# Pre-download models for offline use
python -m anyocr download

# Download models for specific languages
python -m anyocr download --lang en,vi

# Download for a specific backend
python -m anyocr download --backend easyocr --lang en

import anyocr

# Pre-download English models for the default backend
anyocr.download_models(lang=["en"])

# Pre-download for a specific backend and languages
anyocr.download_models(lang=["en", "vi"], backend="easyocr")

Development

git clone https://github.com/vietanhdev/anyocr.git
cd anyocr
pip install -e ".[dev]"
pytest tests/ -v

License

MIT License. See LICENSE for details.

---

Built by Viet-Anh Nguyen | NRL.ai