Project Code Overview

The Spoofing Detection Engine is a software library designed to determine whether a face detected in an image or video stream is genuine or a spoof (a presentation attack, such as showing a photo or a video of a face to the camera). The engine employs a specialized neural network to classify detected faces as either "real" or "spoof," providing a critical layer of security for face-based authentication systems.

Key Components

Core Classes

• SpoofEngine: The main engine class that orchestrates the entire spoofing detection pipeline, from face detection to classification.
• core::face::SpoofClassifier: A dedicated neural network classifier that is the core of the spoofing detection logic. It analyzes a detected face and outputs a score indicating the probability of it being real or a spoof.
• EngineBase: The foundational class that provides a common framework for video processing, handling initialization, pre-processing, and post-processing tasks.
• core::NeuralClassifierBase: A base class for neural network classifiers, providing a standardized interface for running inference. The SpoofClassifier inherits from this class.

Face Detection and Analysis

core::face::FaceDetectorYunet: This class is used for the initial, essential step of detecting faces in the input image or video frame. The detected face region is the input for the spoof classifier.

core::face::FaceMesh: While not directly used for the spoof classification itself, this component can provide detailed facial landmarks which are often used for pre-processing steps like face alignment to ensure consistent input to the classifier.

Supporting Components

• core::Detection: A standard data structure that holds information about a detected face, which is then passed to the classifier.
• core::filter::LandmarkFlowFilter: Useful in video applications to provide temporal stability to face and landmark tracking, ensuring a consistent region is analyzed across frames.

Function Highlights

• Engine Initialization: SpoofEngine::init(resourcePath) initializes the main engine and loads all necessary models, including the weights for the SpoofClassifier.
• Image Processing: SpoofEngine::process(input, output) executes the full pipeline on each frame. It detects faces and then, for each face, runs the spoof classifier to determine its authenticity, visualizing the result on the output image.
• Spoof Classification: core::NeuralClassifierBase::runNetwork(frame, detection) is called by the engine on an instance of SpoofClassifier to get the classification result (real or spoof) for a detected face.
• Face Detection: FaceDetectorYunet::process(frame) finds all faces in the input frame that need to be analyzed.

File and Dependency Structure

The project files are organized in the C:/Projects/Engine/AntalEngine/Engine directory. Key header files include:

• SpoofEngine.h: Contains the declaration of the main engine class.
• SpoofClassifier.h: Defines the neural network classifier for spoofing detection.
• NeuralClassifierBase.h: Provides the base class for the spoof classifier.
• FaceDetectorYunet.h: Provides the core face detection functionality.
• EngineBase.h: Contains the definition for the base engine class.