3D computer vision system for face recognition Ivan A. Matveev, Alexander B. Murynin Computing Center of Russian Academy of Sciences Moscow, Russia Abstract One of the most popular and practically valuable problems in the field of pattern analysis and machine intelligence is human face recognition. A computer vision system for identification of human faces is presented. The system is designed as an automatic checkpoint. Two scenarios of system behavior are developed. The first one assumes verification of personal data, entered by visitor by a card reader. The other performs identification by only visitor’s biometrics. The system performs remote measurements of face features of different types. In addition to stereoscopic images inputted to computer from cameras the models can use voice data and some person physical characteristics such as person’s height, measured by imaging system. The diversity of employed characteristics makes the system reliable and tolerant. Keywords: computer vision, face recognition, 3D vision 1. INTRODUCTION Computer systems for automatic person recognition based on biometrics [1][2] are being developed intensively in connection with creation of computer security systems. Such systems based on recognition of human face and voice have a significant advantage comparing to systems, that use characteristics of fingerprints, iris etc., because recording of face and voice characteristics doesn’t require any physical contact between the person and sensors of the system. It is urgent to create a recognition system, which first, ensures sufficient defense from unauthorized access, and second, implements measurements with minimal discomfort for users. For this purpose it is advisable to make use of methods, based on real time reconstruction of 3-D shape of human face [3]. The usage of 3-D surface increases reliability of the security system because a possibility is eliminated to pass the control by presenting a photo of some other person made in life-size. The system presented contains special equipment and software that is a complex of various image processing and pattern recognition algorithms. Common structure of recognition system based on measurements of human biometrics is described and two laboratory models developed. Different modes of possible system behavior originating from two different types of recognition are presented. 2. SYSTEM STRUCTURE The general structure of the recognition system is illustrated in Figure 1. The system may be treated as a compound of hardware (equipment) and software (algorithms). The authors have developed two laboratory models of system that are denoted here as ‘visible-range’ and ‘infrared’ ones. The software as well can be used in two modes: ‘verification’ and ‘identification’. The detailed discussion of each point will follow and here it would be only emphasized, that any of two software branches can be used with any of hardware models with minimal adjusting of image processing parameters. Visible Infra-red Verification Identification Software Hardware System Figure 1 General system structure Consider hardware structure of person recognition system based on remote measurements of some human biometrics. The model imitates an operation of an automatic entrance checkpoint that accomplishes control of visitors’ access to some object. To carry out this task the system should be able to register data, process it and report results or perform some actions depending on them. Thus system should consist of the following functional blocks: input devices, analyzer block and effectors block. Input devices of the system described contain personal code reader, sound recorder, and video cameras. Analyzer consists of speech recognizer, images analyzer, decision-making algorithm. Effectors may include camera-positioning apparatus and installation like a turnstile for assuring limitations of physical access to a guarded object. Figure 2 sketches a principal scheme of the system. The system functionality can be extended by connecting additional equipment for biometrics measurement, for instance a weight sensor, an apparatus for fingerprint inputting, iris viewing and so on. International Conference Graphicon 1999, Moscow, Russia, http://www.graphicon.ru/