AllerScan

AllerScan is an image-based food safety tool that helps users identify possible allergens before buying or eating food. A user can upload an image of a menu, food label, ingredient list, shelf sign, or product display, and AllerScan uses OCR and allergen detection logic to return possible allergens with confidence scores.

Project Goal

Many people with food allergies have to make quick decisions from incomplete food information. AllerScan is designed to help by combining:

• Optical Character Recognition, or OCR, to extract food names from images
• Rule-based allergen matching using known ingredient keywords
• A multi-label machine learning model that can predict more than one allergen per food item
• CSV output for reviewing and saving allergen prediction results

This project is not intended to replace professional medical advice or official ingredient/allergen labels. It is a support tool for identifying possible risks.

Features

• Upload or process a menu/food image
• Extract text from the image using OCR
• Detect food items from extracted text
• Predict possible allergens for each food item
• Supports multi-label allergen prediction, meaning one food item can have multiple allergens
• Uses rule-based matching first for high-confidence ingredient matches
• Falls back to a machine learning model when direct matches are not found
• Exports prediction results to a CSV file

Repository Structure

AllerScan/
├── data/
│   ├── sample/
│   │   ├── sample_allergen_database.csv
│   │   └── sample_menu.jpg
│   ├── food_ingredients_and_allergens.csv
│   └── training_data.csv
│
├── notebooks/
│   ├── OCR.ipynb
│   ├── foodclassification.ipynb
│   ├── mlmodel.ipynb
│   └── rule_based_matching.ipynb
│
├── src/
│   ├── build_training_data.py
│   ├── predict_allergens.py
│   └── train_allergen_model.py
│
├── models/
│   └── .gitkeep
│
├── outputs/
│   └── .gitkeep
│
├── README.md
├── DATA.md
└── .gitignore