Face recognition system analysis: by the front end face capture system acquisition subsystem, network transmission and back-end subsystem management subsystems resolve to achieve the collection of traffic information face, transmission, processing, analysis and centralized management. System, the front face, head collection devices collect face images, access server main receiving and forwarding information and pictures, can be a variety of models, a number of manufacturers of machine snapshot provides a unified access service, received snapped pictures stored in the cloud storage unit by facial structure analysis server to capture video and image modeling and real-time blacklist than alarm, face information model and data modeling to get the data into large units. Backend analytic application platform is based on the application needs of users, supports real-time facial capture, retrieval and other functions, can provide real-time blacklist database than the snapshot picture of information to the user, provide fast and efficient service to found a suspicious target.
Face detection and image acquisition
Face image capture: different face images can be collected down through the camera lens, such as aspects of still images, moving images, different positions, different facial expressions, etc. can be a good acquisition. When the user who within the shooting range of the collection device, the device will automatically search for and collect the user's face captured image.
Face Detection: face detection in practice is mainly used for face recognition preprocessing, i.e. to calibrate the exact position and size of the face in the image. Mode feature face image contained very rich, such as histogram features, color characteristics, template feature, structure and Haar features and so on. Face detection is useful to pick out where this information and use these features for face detection.
The mainstream of face detection method using Adaboost learning algorithm based on the above features, Adaboost algorithm is a classification method is used, it is some weak classification together, the combination of strong new classification.
Face detection process using the Adaboost algorithm selected some of the most representative feature rectangle face (weak classifiers), the manner of weighted voting weak classifiers configured as a strong classifier, and then a plurality of strong classifiers trained obtained series to form a cascade structure laminated classifier effectively improve the detection speed of the classifier.
Face image preprocessing
Face image pre-processing: preprocessing the image is a face based on face detection results, the image processing service process and ultimately to the feature extraction. An original image acquisition system is limited due to various conditions and random interference, often can not be used directly, it must be gradation correction, noise filtering and other image pre-processing of the image at an early stage. For the face image, its pre-compensation process includes light face image, gradation conversion, histogram equalization, normalization, geometric correction, filtering, and sharpening.
Facial image feature extraction
Facial image feature extraction: feature recognition system may be used usually divided into visual feature, wherein the pixel count, wherein the transform coefficients face image, the face image or the like algebraic features. Facial feature extraction is for certain facial features carried out. Facial feature extraction, also known as the face representation, it is the process of facial feature modeling. Facial feature extraction method summed up into two categories: one is based on the characterization of knowledge; the other is the algebraic features or characterization methods based on statistical learning.
Mainly the shape of a face organ, and describe characteristics of the distance between them contributes to the characteristic data obtained by face knowledge-based characterization methods, which typically comprises a component characterized in Euclidean distance, and angle of curvature between the feature point, etc. . Face consists of eyes, nose, mouth, chin topical, local and geometric description of these structural relationship between them, can be an important feature of face recognition, the features referred to geometric features. Knowledge-based human face representation includes a method and template matching based on geometric features.
Facial image matching and recognition
Face image matching and recognition: feature template extracted face image feature data stored in the database searching and matching, by setting a threshold value, when the similarity exceeds the threshold value, then the output of the matching results obtained. Face recognition is the facial features to be identified have been compared with people face features template, face identity information to judge according to the degree of similarity. This process is divided into two categories: one is confirmed, the process is one image comparison, and the other is to identify, is the process many image matching comparison.