AI-INVOICE-OCR-ENGINE

中文版 | English

A complete AI-powered document OCR (Optical Character Recognition) solution based on PaddleOCR v5, optimized for invoice and accounting document recognition with support for Chinese and English text.

Demo

Chinese Invoice / 中文發票

English Invoice / 英文發票

Architecture Diagrams

1. OCR Inference Pipeline

2. Model Architecture

3. Training Pipeline

4. System Architecture

5. Data Flow

6. Deployment Architecture

7. Database ER Diagram

Features

• High Accuracy: Utilizes PP-OCRv5 server models for superior recognition quality
• Multi-language Support: Chinese, English, and mixed text recognition
• Document Preprocessing: Automatic orientation correction and document unwarping
• Bounding Box Visualization: Draw green boxes around detected text regions
• Text Export: Save recognition results to TXT file
• Easy Integration: Simple Python API for quick deployment

Architecture Overview

Input Image → Preprocessing → Text Detection → Text Recognition → Output
                                    ↓                  ↓
                              Green Boxes         TXT File

Pipeline Components

Model Architecture

• Text Detection (DB Algorithm)

  • Backbone: ResNet50_vd
  • Neck: Feature Pyramid Network (FPN)
  • Head: Differentiable Binarization with probability, threshold, and binary maps

• Text Recognition (SVTR)

  • Patch Embedding with positional encoding
  • 4-stage encoder with local and global mixing
  • CTC (Connectionist Temporal Classification) decoder
  • Vocabulary: 6,624 characters

Installation

Requirements

• Python 3.8+
• Windows/Linux/macOS

Install Dependencies

pip install paddlepaddle>=2.5.0
pip install paddleocr>=3.0.0
pip install pillow>=9.0.0
pip install opencv-python>=4.5.0

Quick Start

Basic Usage

from paddleocr import PaddleOCR

# Initialize OCR engine
ocr = PaddleOCR(
    use_doc_orientation_classify=False,
    use_doc_unwarping=False
)

# Perform OCR on an image
result = ocr.predict('invoice.png')

# Process results
for item in result:
    if 'rec_texts' in item:
        for i, text in enumerate(item['rec_texts']):
            score = item['rec_scores'][i]
            print(f"Text: {text}, Confidence: {score:.2%}")

Draw Boxes Only (No Text on Image)

import cv2
import numpy as np

def draw_boxes_only(image_path, boxes, output_path):
    img = cv2.imread(image_path)
    for box in boxes:
        pts = np.array(box).astype(np.int32)
        cv2.polylines(img, [pts], True, (0, 255, 0), 2)
    cv2.imwrite(output_path, img)

Save Results to TXT

def save_text_to_file(txts, scores, output_path):
    with open(output_path, 'w', encoding='utf-8') as f:
        for i, (txt, score) in enumerate(zip(txts, scores)):
            f.write(f"{i+1}. {txt} (Confidence: {score:.2%})\n")

Project Structure

AI-INVOICE-OCR-ENGINE/
├── README.md
├── README_CN.md
├── LICENSE-MIT
├── setup.py
├── demo/
│   ├── invoice_ch.png           # Chinese invoice (input)
│   ├── invoice_en.png           # English invoice (input)
│   ├── result_invoice_cn.jpg    # Chinese result (output)
│   ├── result_invoice_en.jpg    # English result (output)
│   ├── 1.png ~ 7.png            # Architecture diagrams
│   └── *.puml                   # PlantUML source files
├── models/
│   └── pretrained/
│       └── weights/             # Model files (.onnx, .pdmodel, .pdiparams)
├── preprocess/
├── detect/
├── recognize/
├── data/
├── losses/
├── train/
└── utils/

Performance

*Times measured on Intel i7-10700 CPU

Supported Document Types

• Invoices and receipts
• Financial statements
• Contracts and agreements
• ID cards and certificates
• General printed documents
• Mixed Chinese/English documents

Author

Jammy Lin Email: [email protected]

License

This project is licensed under the MIT License - see the LICENSE-MIT file for details.

Acknowledgments

• PaddleOCR - The underlying OCR engine
• PaddlePaddle - Deep learning framework

References

• PP-OCRv5: PaddleOCR Documentation
• DB Algorithm: Real-time Scene Text Detection with Differentiable Binarization
• SVTR: SVTR: Scene Text Recognition with a Single Visual Model

---

⭐ Support This Project

If this project helps you, please give it a Star!

Your support is my motivation to keep improving!

Ways to support:

• ⭐ Star this repository
• 🍴 Fork and contribute
• 🐛 Report issues and suggestions
• 📢 Share with others!