Abstract—This paper proposes the implemental of a Multi- biometric face recognition system based on levels of fusion using stereo images. Only three fusion levels are used for developing the proposed face recognition system which are; sensor level fusion, feature level fusion and decision level fusion; where for each level of fusion was used the Eigenfaces and Gabor Filters for the feature extraction while for data classification the back propagation neural network and Support Vector Machine are used. At each level of fusion it is possible to evaluate the performance of proposed face recognition marking the highest identification and verification rate. Giving the best result the feature level fusion that consist of two images with different angles of face, because with this we get more coefficients or information that constitute a new template of the face. This provides a higher recognition rate in comparison with the systems that use a picture with a single angle. Also was observed the system behavior with different features extraction and classification algorithms. Evaluation results show that the best results are obtained using the Gabor filters with the Support Vector Machine, because in this case the recognition rates are higher and with less computation time. Keywords—Multi-biometric system, levels of fusion, Eigenfaces, Gabor filters, Neural Network Back propagation, Support Vector Machine. I. INTRODUCTION biometric system is essentially a pattern recognition system that acquires biometric data from an individual, extracts a salient feature set from these data, compares these features set with the feature sets stored in the database and executes an action based on the result of such comparison. These systems perform, basically, two tasks: identification and identity verification [1]. In the first one the goal is to determine the person most likely to be among a set of persons, while in the second one to verify if the person is who he/she claims to be, in both bases, based on the physical (face, Elizabeth Garcia-Rios is enrolled in the PhD Program in Communications and Electronics Engineers, Instituto Politecnico Nacional, Mexico D. F. Mexico, e-mail: (egarciar1009@ alumno.ipn.mx). Enrique Escamilla-Hernandez is with the Instituto Politecnico Nacional, Mexico D. F., Mexico, (e-mail: eescamillah@ipn.mx) Gualberto Aguilar-Torres is with the Instituto Politecnico Nacional, Mexico D. F. Mexico, e-mail: (autg79y@yahoo.com) Omar Jacobo-Sanchez is with the Universidad Autonoma de Hidalgo, Mexico, (email zecmol@hotmail.com) Mariko Nakano-Miyatake is with the Instituto Politecnico Nacional, Mexico D. F. Mexico, e-mail: (mnakano@ipn.mx) Hector Perez-Meana is with the Instituto Politecnico Nacional, Mexico D. F. Mexico, (e-mail: hmperezm@ipn.mx). fingerprint, iris, voice, etc.) or behavioral features (signature, dynamic typing, way of walking, etc.) of the people under analysis [1], [2]. Among the biometric systems, the face is one of the most widely used, because it is a non-intrusive method and where the facial attributes, like; eyes, eyebrows, nose, lips and skin, which are the main features used to make a personal recognition are capture by a camera [3]-[9]. Face recognition has received significant attention in the last several years because it is increasingly being used in many commercial applications due to the rapid electronic technology evolution. As a result several efficient algorithms have been proposed during the last two decades and many of them are even currently available on the market. However almost all still require some special conditions for its optimal operation such as: the pose and orientation of face, background lighting and diverse face expressions. Also the quality of input image is very important because, if the image face is noisy, poorly captured or presents large occluded regions, the subsequent process may be severely affected and as a consequence, the recognition rates may become lower. Therefore it is important to obtain an input image containing as much information as possible in order to have a better assessment of the face characteristics [10] Other limitations of most face recognition systems, which has received little attention, is the fact that the face is a three dimensional object and most currently proposed face recognition systems are based on two dimensions images [4]- [9] which may be cheated by placing a photo of another person in front of the input camera. To avoid this problem, this paper proposes a face recognition system which uses stereo images of face, doing the fusion of this information, for improving the precision of a face recognition system. At entrance, the system captures stereo images which are firstly used to determine in the captures images corresponds to a three dimensional object or a two dimensional picture. Next different three different levels of fusion as analyzed such as: sensor level fusion, feature level fusion and decision level fusion. The usage of these levels of fusion is with two components of stereo image in an individual form, combining them to form a 3-dimensional face model. To develop the feature extraction stage two efficient methods are used: the eigenfaces [3], [7] and Gabor filters [7]-[9] methods. Finally in the classification stage the methods are analyzed: the Back-propagation neural network [1], [7], [11] and the Support Vector Machine [9]-[15]. The experimental results show that the proposed multi-biometric system overcomes some of the limitations present in an Multi-biometric Face Recognition System using Levels of Fusion. Elizabeth Garcia-Rios, Enrique Escamilla-Hernandez, Gualberto Aguilar-Torres, Omar Jacobo-Sanchez, Mariko Nakano-Miyatake and Hector Perez-Meana A INTERNATIONAL JOURNAL OF COMPUTERS Issue 3, Volume 7, 2013 99