Skip to content
All projects
AI & Machine LearningPrivate

Biometric Authentication & ML.NET Client

A secure desktop client integrating real-time OpenCV facial bounding, ML.NET deep learning inference, and Firebase for biometric two-factor authentication.

Tech stack

Languages
Frameworks & libraries

Performance & metrics

  • Strict memory lifecycle management of unmanaged OpenCV Mat and VideoCapture objects via explicit Dispose blocks to prevent memory leaks during continuous webcam streaming
  • Low-latency desktop inference achieved by statically compiling the ML.NET prediction engine directly into the application binary

Why I built it

Developed during an internship to solve the vulnerabilities inherent in traditional static-credential systems. I needed a dynamic pipeline that validates both cryptographic password hashes and real-time spatial biometrics before granting system access. To train the underlying neural network, I utilized datasets autonomously extracted via my proprietary C++ OpenCV-Face-Features-Toolkit-For-ML utility.

Architecture & decisions

The application is engineered on .NET 6.0. Real-time video capture and spatial face bounding are offloaded to OpenCvSharp leveraging Haar Cascades to minimize CPU overhead during the detection phase. Isolated facial matrices are passed into an embedded ML.NET computational graph—a pre-trained ResNet50 Deep Neural Network—for hardware-accelerated classification. State synchronization and user registry validation are managed asynchronously via Firebase Realtime Database (FireSharp), enforcing SHA-256 cryptographic hashing for credential storage.

Key features

  1. 01Real-time video capture and spatial facial isolation utilizing OpenCvSharp native bindings
  2. 02Embedded ML.NET inference pipeline utilizing a ResNet50 DNN architecture for high-accuracy classification
  3. 03Asynchronous Firebase Realtime Database integration for distributed user registry management
  4. 04Biometric Two-Factor Authentication (2FA) combining SHA-256 credential hashing with localized neural network validation