A production-grade OCR application designed for the Fintech sector. This project demonstrates advanced on-device text recognition for credit/debit cards and bank passbooks, utilizing a heuristic-based parsing engine for high accuracy in real-world conditions.

• Details Extracted: Card Number, Expiry Date, Card Holder Name.
• Privacy: Automated masking of the card number (e.g., XXXX XXXX XXXX 1234) with a toggle to view.
• Algorithm: CardParser manually processes noisy OCR text using a scoring system.
• Luhn Validation: Manual implementation of the Luhn Algorithm (LuhnValidator) to verify card authenticity.

• Details Extracted: Account Holder Name, Account Number, IFSC Code.
• Algorithm: PassbookParser uses keyword proximity and heuristic scoring to identify account details amidst transaction history and noisy text.
• Normalization: Standardizes Indian IFSC codes (ensuring the 5th character is always 0).

As per assignment constraints, no external libraries were used for parsing the extracted OCR text.

1. Candidate Identification: Scans text for 13-19 digit sequences.
2. Prioritization: Ranks candidates based on Luhn validity and standard card lengths.
3. Expiry Detection: Searches for MM/YY or MM/YYYY patterns. Scores them higher if near keywords like "Valid Thru" or "Expiry" and lower if near "DOB".
4. Name Extraction: Identifies 2-3 word uppercase strings excluding card brand names and banking keywords.

1. IFSC Detection: Regex-based pattern matching for standard 11-digit IFSC codes.
2. Account Number Extraction: Identifies numeric sequences of 9-18 digits. Uses a scoring system where proximity to "A/C" or "Account" increases confidence, while proximity to "Balance" or "Amount" decreases it.
3. Name Extraction: Looks for name-related keywords and extracts the following alphabetic string.

• Language: Optimized for English-language documents.
• IFSC Format: Assumes the standard 11-character Indian IFSC format.
• Card Holder Name: Extracted if printed in standard uppercase format on the card front.
• Image Quality: Assumes reasonable lighting and focus for on-device OCR accuracy.

• Bank Logo Recognition: While branding helps scoring, a full logo-to-bank mapping was skipped to focus on text-based parsing.
• Non-Standard Expiry: Very rare formats (e.g., text-based "Dec 2025") were deprioritized over numeric formats.
• Offline ML Models: Used Google ML Kit which is on-device but requires initial library download; fully air-gapped custom Tesseract models were skipped for better performance/size ratio.

• State Management: Cubit (Flutter Bloc) for predictable, testable business logic.
• Dependency Injection: GetIt for service decoupling.
• OCR Engine: Google ML Kit Text Recognition.
• Navigation: GoRouter for declarative routing.
• Architecture: Feature-first Clean Architecture.

lib/
 ├── core/          # Shared utilities (Regex, Luhn, Text Normalization)
 ├── features/      # Domain-specific logic (Card, Passbook)
 ├── shared/        # Reusable UI components
 └── app/           # DI, Theme, and Routing configuration

1. Clone the repository

   git clone https://github.com/codebyforam/flutter-scanner.git

2. Install Dependencies

   flutter pub get

3. Run Tests

   flutter test

4. Run Application

   flutter run

codeByForam

• GitHub: @codeByForam

Distributed under the MIT License. See LICENSE for more information.