face identification

The existing security systems, despite large saturation with cameras and continuous recording of video, do not enable active identification and verification of persons. To provide this functionality, POLIXEL S.A. developed for M3S system, its proprietary method of face recognition, which may be used to create compact face representations, which may be stored e.g. on chip cards. High effectiveness of this method both in person identification and verification has been proved in tests.

Face recognition technologies have two main applications: verification and identification. In the first case, person database is usually quite limited in size, and includes contains characteristics of person faces in addition to typical personal information and data used by classic verification or access control system. Using recognition terminology, these face characteristics are called face descriptors. They provide a compact representation of a person's face, what allows for effective storage of large databases of identified persons, and fast comparisons. Verification based on face recognition plays a supplementary role in person verification systems, but is very important because of its very high effectiveness reaching 100% and allowing sure identification. It may be employed for example to prevent unauthorised use of stolen ID or payment cards.

Identification is much more complicated, as it involves one-to-many and many-to-many matching relations. It is utilised in particular in the systems monitoring large premises (airports, railway stations, administration buildings, etc.), with many cameras controlling simultaneously many locations. Person database used for identification is usually much larger than verification databases, and includes descriptors (optionally also pictures) of persons wanted by the police or undesirable in specific locations.

Recognition of faces, or broadly speaking of visual objects, must be preceded by a painstaking process of face localisation in video pictures, face normalisation, feature filtering. These operations play a supplementary role in the recognition process, but their impact on final results is very significant. Many methods of face localisation have been described in literature, with their effectiveness depending on a complexity of analysed picture and intensity of movement between sequential frames. As recognition algorithm is required to integrate with monitoring system, the solution based on motion estimates was chosen. It locates objects within analysed sequence of video pictures. After segmentation, pictures are subject to local feature analysis, intended to identify face representations.

The above descriptor-based method, enables storage of compact face descriptors on chip cards, which may be used to verify identity of a person by a recognition system equipped with a camera and suitable software.