This is a complete face-authentication system.
I developed it during my degree-thesis (italian language) in Computer Engineering. It is entrely written in C++. The whole computative process mainly consists of the following functional modules:
face-tracking
OpenCV classifier + template matching technique for angled head-posture detection and a continuos tracking without any visual freeze.
eyes-tracking
for face-cropping (missing a complete face landamaks detection system...) and geometric normalization
(a planar face rotation to normalize user's head-posture)
brightness-normalization
For reducing environment light condition variability (the system works with just a RGB webcam)
face-antispoofing
Multiresolution/Multiscale LBP and SVM with radial filter classification.
The system is able to work without any kind of liveness-detection, so it doesn't need user cooperation and it's
absolutely transparent for user.
features-extraction
The features-extraction module is based on the Local Quantized Pattern methodology (LQP), with vector
quantization (clustering) performed with k-means algorithm (scratch programmed).
The system is provided of a GUI for helping user during the enollment and the classification phases of the authentication process.
Windows/Linux OS compatibility. Third party library involved: OpenCV 2.4.9, Qt 5.1.
System divided in a Trainer application and a Recognizer application.
More info at: vcolamatteo/computervision.
For any info request email me at: [email protected]