The facial recognition system consists of four main components: facial image acquisition and detection, facial image preprocessing, feature extraction, and matching and recognition.

Facial Image Acquisition and Detection

Facial Image Capture: Different types of facial images can be captured through the camera lens, including static and dynamic images, various positions, and different expressions, all with excellent results. When a user is within the shooting range of the capture device, the device automatically searches for and captures the user's facial image.

Facial Detection: Primarily used in the preprocessing of facial recognition, facial detection accurately locates and sizes the faces in images. The pattern features within facial images are abundant, including histogram features, color features, template features, structural features, and Haar features. Facial detection involves extracting this useful information and utilizing these features to perform facial detection.

The mainstream face detection methods utilize the aforementioned features and adopt the Adaboost learning algorithm. Adaboost is a classification method that combines several relatively weak classification techniques to create a new, highly effective one.

During the face detection process, the Adaboost algorithm is used to select some rectangular features that represent faces (weak classifiers), which are then combined into a strong classifier through a weighted voting method. Several strong classifiers trained are linked together to form a cascading structure of a layered classifier, effectively enhancing the detection speed of the classifier.

Facial Image Preprocessing

Facial Image Preprocessing: The pre-processing of facial images is based on the detection results, which involves processing the images to serve the feature extraction process. The original images obtained by the system are often unusable directly due to various constraints and random disturbances, and they must undergo image pre-processing such as grayscale correction and noise filtering in the early stage of image processing. For facial images, the pre-processing process mainly includes light compensation, grayscale transformation, histogram equalization, normalization, geometric correction, filtering, and sharpening.

Facial Image Feature Extraction

Facial Image Feature Extraction: The features typically used in facial recognition systems are usually categorized into visual features, pixel statistical features, facial image transformation coefficient features, and algebraic facial image features, among others. Facial feature extraction specifically targets certain features of the face. Facial feature extraction, also known as facial representation, is the process of building a feature model for the face. The methods of facial feature extraction can be summarized into two main categories: one is a knowledge-based representation method; the other is a representation method based on algebraic features or statistical learning.

Knowledge-based representation methods primarily obtain facial classification feature data based on the shape descriptions of facial organs and their distance characteristics. The feature components usually include Euclidean distances between feature points, curvatures, and angles. The face is composed of local parts such as eyes, nose, mouth, and chin. The geometric descriptions of these local parts and their structural relationships can serve as important features for facial recognition, known as geometric features. Knowledge-based facial representation mainly includes methods based on geometric features and template matching.

Facial Image Matching and Recognition

Facial Image Matching and Recognition: The extracted facial image feature data is searched and matched against stored feature templates in the database. By setting a threshold, results are output when the similarity exceeds this threshold. Facial recognition involves comparing the features of a face to be identified with existing facial feature templates and determining the face's identity based on the similarity level. This process is divided into two categories: confirmation, which is a one-to-one image comparison process, and identification, which is a one-to-many image matching and comparison process.