Skip to content

README.md

Latest commit

 

History

History
90 lines (69 loc) · 2.79 KB

README.md

File metadata and controls

90 lines (69 loc) · 2.79 KB
Raw

POLY-SCAN-IDs

A Streamlit-based web application that automatically extracts structured information from various ID documents using OCR, deep learning, and face detection models. The app supports national ID cards, driver's licenses, and passports in French.

Features

  • Upload and process scanned images of ID documents
  • Advanced OCR using Doctr
  • Automatic document type classification (ID front, ID back, license, passport)
  • OCR filtering (French and Arabic)
  • Face detection and cropping
  • Download extracted data as JSON
  • Download extracted face image
  • Clean, styled UI

Supported Document Types

  • Moroccan National Identity Cards (Front & Back)
  • Moroccan Driver's Licenses
  • Moroccan Passports

Tech Stack

Technology Description
Streamlit For building the interactive web UI
Doctr (OCR) For text detection and recognition
PyTorch For the language classifier model
OpenCV Image processing
YOLO Face detection
PIL (Pillow) Image enhancements
Custom Modules For document-specific parsing and layout

Sample Workflow

  1. Upload an image of a document (.jpg, .jpeg, or .png)
  2. Click the "Extract Information" button
  3. View:
    • The document with detected text overlays
    • Detected face (if present)
    • Parsed personal data displayed in styled card format
  4. Download the:
    • Extracted JSON data
    • Cropped face image (if available)

Project Structure (Modules)

Module Purpose
craft_ocr.py Text box detection using CRAFT or similar model
id_frt.py Extraction logic for front of national ID
id_bck.py Extraction logic for back of national ID
perm_con.py Extraction logic for driver's licenses
passp.py Extraction logic for passports
type_classf.py Document type classification (Model training)
face_dtc.py Face detection logic using YOLO

Models Used

Model Purpose File Path
doctr_model.pth OCR predictor /models/doctr_model.pth
language_classifier.pth Classify OCR language (Arabic vs French) /models/language_classifier.pth
yolov8x.pt Face detection model /models/yolov8x.pt (Or auto-Download)

Tips for Best Results

  • Use high-resolution images
  • Ensure good lighting and minimal glare
  • Place the document on a dark, non-reflective background
  • Avoid skewed or rotated documents (Important)

Installation

You'll need Python 3.8+ installed

# 1. Clone the repository
git clone https://github.com/mouradboutrid/poly-scan-IDs.git
cd id-document-extractor

# 2. (Optional) Create virtual environment
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# 3. Install dependencies
pip install -r requirements.txt

# 4. Run the Streamlit app
streamlit src/run app.py