The Invoice Parsing App is a mobile application designed to enhance the efficiency of invoice scanning and data extraction. Utilizing lightweight, on-device machine learning models, the app provides a robust offline tool for parsing and managing invoices. This ensures enhanced privacy and rapid inference without reliance on server infrastructure.
- 🔍 OCR-Based Scanning: Extracts key information such as supplier details, itemized purchases, and taxes.
- 🤖 Hybrid Model Integration:
- 🛠️ ML Kit's Entity Extractor: Identifies general fields like dates, addresses, and emails.
- 📊 Fine-Tuned BiLSTM NER Model: Extracts domain-specific entities such as invoice numbers and supplier names.
- 📐 Proximity-Based Detection: Recognizes entities like GST numbers and invoice IDs using spatial relationships.
- ✏️ Editable User Interface: Users can verify and manually correct extracted data.
- 💾 Local Storage: Parsed data is stored offline to enhance privacy and accessibility.
- A dataset of 1,000 Tally invoices was collected and annotated.
- Regex patterns were employed to label entities such as dates, amounts, and invoice numbers.
- A pre-trained BiLSTM NER model was fine-tuned for improved performance.
- The model was optimized for on-device usage using TensorFlow Lite (TFLite).
- Results from ML Kit and BiLSTM were combined into a unified system.
- Supported local Datastore DB for structured offline data management.
- 🌀 Complex Layouts: The app struggles with unconventional invoice layouts.
- 📄 Preprocessing Dependency: Requires high-quality scans for accurate extraction.
- 🤷♂️ Model Conflict Resolution: Manual verification is necessary for conflicting outputs.
- 📊 Dataset Limitations: Performance is tied to the quality of the training dataset.
- 🌐 Multi-Language Support: Plans to expand capabilities to handle invoices in multiple languages.
- 📱 Cross-Platform Compatibility: Extend support to iOS for broader accessibility.
- 📚 Batch Processing: Enable simultaneous processing of multiple documents.
- 📑 Additional Document Types: Incorporate support for PDFs, purchase orders, and contracts.
- 🖥️ Android Studio (for development and testing)
- 🐍 Python (for dataset preparation and model training)
- 🤖 TensorFlow Lite (for on-device model optimization)
You can download the APK file for the Invoice Parsing App using the link below:
Test app on this invoice: PerfectVisionInvoice_2024-07-08_18-33-16_45.pdf
Watch a demo of the app in action:
To get started with the Invoice Parsing App, clone the repository using the following command:
git clone https://github.com/CulturalProfessor/invoice-parsing-app.git