Face Recognition System
Categorization and Features
Face recognition system is a biometric system-any automatically measurable, robust and distinctive traits that can be used to identify an individual- that operates in two modes; face verification or authentication, and face identification or recognition (Li & Jain, 2011). Face verification postulates a one to one match that likens a face image in question against an enrollment face image whose identity is being claimed (Li & Jain, 2011). A distinctive application of this mode is person verification for self served immigration clearance using E-passport (Li & Jain, 2011). On the other hand, face identification encompasses one-to-many matching that compares a face in question against multiple faces, in the enrollment database, in order to associate the identity of the query face to one of those in the data base (Li & Jain, 2011). An application of this mode may include the need to recognize a most similar face, in either a watch list check or face identification in a surveillance video (Li & Jain, 2011). In addition to this, the face identification mode incorporates a confidence level threshold under which, in searching for a face in query, all the faces whose similarity scores are above the given threshold will be reported, for a proficient identification of the face in query (Li & Jain, 2011).
In light with this, there is a plethora of facets that determine adept operation of the face recognition system, and they include illumination, facial pose, age span, hair, facial wear, expression, and motion (Li & Jain, 2011). Further, the face recognition system application can be categorized in terms of a user’s cooperation; cooperative user scenarios, and non-cooperative user scenarios (Li & Jain, 2011). Cooperative application is highly depicted in incidences like physical access control, computer login, and E passport. In each and every incidence, the user is amenable to cooperate, and comply with requests of the system, by availing the face in a proper way, in order to be accorded access or the expected privileges (Li & Jain, 2011). Besides, non-cooperative scenarios involve the application of surveillance, and the user is unaware of the process of identification (Li & Jain, 2011).
Software Used in Face Recognition System
There are varieties of software used in the establishment of the face recognition system. Facelt software is a facial recognition software that enables rapid and accurate detection or recognition of faces by the computer (Newman, 2009). The software allows a broad range of applications, which embraces face detection, face recognition, image quality, segmentation, face print, tracking and compression (Newman, 2009). Video image Retrieval and Analysis Tool (VIRAT) is also a crucial program used in face recognition system. It allows large storage of video, which allows provision of adequate information that can be used in matching task. In addition to this, animation software can also be used in face recognition systems, so as to make face identification simple and fast.
How the System Works
The system always involves the principle of face recognition algorithms, which comprise of visual pattern recognition problem (Li & Jain, 2011). The face recognition system consists of four modules, which are crucial in the face recognition process. They comprise of face detection, normalization, feature extraction, and matching (Li & Jain, 2011). In face detection, the computer is to determine if the pixels in the image are part of the face in the query and which are not (Introna & Nissenbaum, n. d). The process becomes easier when identifying a face from a passport, where the background is visible and clear. However, when the background is littered up with objects, the problem becomes complex (Introna & Nissenbaum, n. d). In tandem to this, face detection also offers a provision on coarse estimation of the location and the scale of the face. It also localizes facial landmarks; eyes, noses, mouth and facial outline (Li & Jain, 2011). Further, the facial landmark can be accomplished by a land marking or facial alignment module (Li & Jain, 2011).
Face normalization is performed to normalize the face, which entails standardization of the detected face, geometrically and photometrically (Li & Jain, 2011: Introna & Nissenbaum, n. d). The geometrical normalization involves transformation of the face into a standard frame through face cropping. Moreover, warping and morphing are also incorporated in detailed or more elaborate geometric normalization (Li & Jain, 2011). In line with this, Gray scale and illumination aspects play a vital role in photometric normalization (Li & Jain, 2011). Also, in the normalization process, there is accurate location of cardinal facial landmarks, which are the key to all systems, irrespective of all methods of recognition used (Introna & Nissenbaum, n. d). Conversely, this process is extremely essential because the recognition methods are expected to recognize face images with varied poses and illumination (Li & Jain, 2011).
The third module is face feature extraction, performed on the normalized face, and it involves the generation of a biometric reference or template (Li & Jain, 2011). The generated biometric template, which are mathematical, presentations are remarkably indispensable in extracting salient information useful in distinguishing faces, in accordance to photometric and geometric variations (Li & Jain, 2011). The templates are then stored in the database to perform any recognition task (Introna & Nissenbaum, n. d). The last process is the face matching, which involves matching of the extracted face against one or many extracted faces available in the database. The core challenge that presents itself in this stage of face recognition is the allocation of suitable metric for the comparison of facial features (Li & Jain, 2011).
After the face recognition process, there is an analysis of the image extracted, using the face recognition algorithms (Lu, n. d). Taking the algorithms into consideration, the operation of the system in terms of image identification can be classified in two; appearance based method and model-based schemes (Lu, n. d). Appearance based method has been the most dominant technique used in the analysis for the past decade, and it comprises presentation of linear or subspace analysis schemes, and non linear or manifolds analysis approaches for face recognition (Lu, n. d). In light with this, appearance based method operates directly on the pixel intensities, and it is used either in a holistic or a local way (Li & Jain, 2011). The holistic method identifies a face using an input as a vector that represent the whole face image, and the local method uses the information of the face image available on localities (Li & Jain, 2011).
In the presentation of linear subspace analysis schemes, the local method is developed into various categories, which encompass Principal Component Analysis (PCA), Independent Component Analysis (ICA), and Linear Discriminant Analysis (LDA) (Jain, Flynn & Ross, 2008). PCA is an eigenface algorithm that reduces dimensionally for the identification of the vector that best accounts for the distribution of the face images. ICA, is similar to PCA except that for ICA, the components are made in non- Gaussian form. LDA uses fisherface algorithm that applies particular information (Jain, Flynn & Ross, 2008). In non linear or manifolds, kernel principal component analysis is used, and it accounts for variations in facial appearance (Li & Jain, 2011).
Conventionally, model based face recognition scheme is designed to construct a model of the face in query (Jain, Flynn & Ross, 2008). The model is classified into broad categories; 2D, which comprises of elastic bunch graph matching (EBGM), and active appearance model (AAM), and 3D, which comprises of 3D morphable model (Jain, Flynn & Ross, 2008). The EBGM constructs dynamic link architecture using image graphs to represent the individual faces (Jain, Flynn & Ross, 2008). Besides, AAM is a statistical model that interprets facial images with known parameters (Jain, Flynn & Ross, 2008). In 3D morphable model, the human face lies in 3D space, thence making it easier to handle the facial variations (Lu, n. d). After the analysis, the encrypted face is then set for performance evaluation and final recognition task, which involves the confirmation of the match between the face print obtained and the face in the database.
Advantages and Disadvantages of Facial Recognition System
The utmost security advantages of using biometrics to verify identification includes the following; elimination of the misuse of lost or stolen cards, it also allows the replacement of Personal Identity Numbers (PINs) with biometric characteristics(Li & Jain, 2011). This allows a convenient and secure access to financial and computer applications (Li & Jain, 2011). In tandem to this, efficiency is also increased in areas or situation where access controls to buildings is automated (Li & Jain, 2011). Moreover, the face recognition system also avails temporal information, which is decidedly crucial in tracing faces, it also stores abundant data in that there are more than enough frames for a certain face available for the recognition task. The core disfavors are that the systems are prone to errors, since they handle and encrypt a lot of data. In conjunction to this, the 2D model can also be highly affected with illumination and any slight alteration in an individual face variation. At some point, it may also be difficult in drawing clear cut difference between identical twins.
Application of face recognition system
According to Li and Jain (2011), the face recognition system can be applied in various categories; in face identification for instance in driver licenses, entitlement programs, passports, national ID, and immigration. It is also used in access control in scenarios like border crossing control, facility, vehicle, computer and database access (Li & Jain, 2011). Moreover, it is applied in security and surveillance, in establishing secure flight boarding system, stadium audience scanning, computer security, also in terrorist alert (Li & Jain, 2011). Other applications may entail; application in law enforcement through crime stopping and suspect alert, suspect tracking, and shoplifter recognition, face databases, multimedia management, human computer interaction and in smart cards (Li & Jain, 2011).
Conclusion
Concisely, face recognition system, is a fundamental system that can aid in improving the security and surveillance system of places; airports, shopping centers, stadiums, and offices. Similarly, it makes operations basic and easy for instance, in face identification. The features of the face recognition system are also intelligible except of a few complications encountered in encrypting the obtained face. The management of both the hard and software can also lead to an establishment of a proficient face recognition system. Nevertheless, they are few problems associated with the system, attributed to the large number of data intake; thus it can amount to errors and generation of false information. With the increasing trends in technology, the errors can be minimized, and systems can work more efficiently.
References
Introna, D. L. & Nissenbaum, H. (n. d). Facial Recognition Technology: A Survey of Policy and Implementation Issues. Retrieved from
http://www.nyu.edu/ccpr/pubs/Niss_04.08.09.pdf
Jain, K. A., Flynn, J. P. & Ross, A. A. (Eds.). (2008). Handbook of Biometrics. New York, NY: Springer Science + Business Media, LLC.
Li, Z. S. & Jain, K. A. (Eds.). (2011). Handbook of Face Recognition (2nd Ed.). London: Springer.
Lu, X. (n. d). Image Analysis for Face Recognition. Retrieved from
http://www.face-rec.org/interesting-papers/General/ImAna4FacRcg_lu.pdf
Newman, R. (2010). Security and Access Control Using Biometric Technologies. Boston, MA: Cengage Learning.
