July 28, 2023
In recent years, with continuous development of security industry, intelligence has become historical trend of future development of security industry. Identification is a major security issue. Due to needs of counter-terrorism, national security and social security, intelligent technology in security industry is also receiving more and more attention. In this context, among various biometric technologies, facial recognition technology stands out with advantages of non-contact, unobtrusiveness, friendliness, intuitiveness, speed, simplicity and scalability. So what is principle behind facial recognition technology?
Face recognition is mainly divided into three processes: face detection, feature extraction, and face recognition.
Face recognition. Face detection is detection and extraction of face images from input images, typically using haar functions and Adaboost algorithm to train cascade classifiers to classify each image block. If some rectangular region passes cascade classifier, it is identified as a face image.
Feature Extraction. Feature extraction means representing information about faces with some numbers, and these numbers are features we want to extract. General facial features fall into two categories: geometric, representative. Geometric features refer to geometric relationships between facial features such as eyes, nose, and mouth, such as distance, area, and angle. Since algorithm uses some intuitive functions, amount of computation is small. However, its scope is limited, since it is impossible to accurately select required characteristic points. In addition, when illumination changes, face is covered with foreign objects, facial expression changes, and features change greatly. Therefore, this type of algorithm is only suitable for coarse face image recognition and cannot be applied in practice.
Face recognition. The face recognition mentioned here is face recognition in a narrow sense, that is, comparison of features extracted from a face for recognition with facial features in a database, and according to a discriminant similarity classification. And face recognition can be divided into two categories: one is confirmation, that is, process of comparing an image of a face with an image of a person already stored in database and answers question of whether you are you; other is identification, which is process of matching an image of a face with all images already stored in database, answering question of who you are. Obviously, face recognition is more difficult than face confirmation because recognition requires massive data to be matched.
Similar to application of fingerprints, relatively mature facial recognition technology is also an attendance machine. Because in attendance system, user is actively collaborating and can get a face that meets requirements in a particular environment. This provides a good source of input for face recognition and can often produce satisfactory results. However, due to problem of light and angle of CCTV sensors installed in some public places, it is difficult to successfully compare received images of faces. This is also one of tasks that need to be solved with development of face recognition technology in future.
