J.A. Carrasco-Ochoa et al. (Eds.): MCPR 2013, LNCS 7914, pp. 114–125, 2013. © Springer-Verlag Berlin Heidelberg 2013 Video Images Fusion to Improve Iris Recognition Accuracy in Unconstrained Environments Juan M. Colores-Vargas 1 , Mireya García-Vázquez 1 , Alejandro Ramírez-Acosta 2 , Héctor Pérez-Meana 3 , and Mariko Nakano - Miyatake 3 1 Instituto Politécnico Nacional, CITEDI Avenida del Parque 1310, Tijuana, B.C. México 22510 2 MIRAL R&D, Palm Garden, Imperial Beach, USA 91932 3 Instituto Politécnico Nacional, ESIME-Culhuacan, DF., México colores@citedi.mx, msarai@ipn.mx, ramacos10@hotmail.com, {mnakano,hmperezm}@ipn.mx Abstract. To date, research on the iris recognition systems are focused on the optimization and proposals of new stages for uncontrolled environment systems to improve the recognition rate levels. In this paper we propose to exploit the biometric information from video-iris, creating a fusioned normalized template through an image fusion technique. Indeed, this method merges the biometric features of a group of video images getting an enhanced image which therefore improves the recognition rates iris, in terms of Hamming distance, in an uncontrolled environment system. We analyzed seven different methods based on pixel-level and multi-resolution fusion techniques on a subset of images from the MBGC.v2 database. The experimental results show that the PCA method presents the best performance to improve recognition values according to the Hamming distances in 83% of the experiments. Keywords: Fusion, Iris, MBGC, PCA, Recognition. 1 Introduction To date, the commercial iris recognition systems based on still images [1-2] are designed to work with special or restricted conditions. This means that they require an ideal environment and cooperative user’s behavior during the iris image acquisition stage to obtain high quality images. Therefore, if any of these requirements are not met, it can cause a substantially increase of error rates, specially the false rejections. Many factors can affect the quality of an iris image, including defocus, motion blur, dilation and heavy occlusion. Naturally, poor image’s quality cannot generate satisfactory recognition because they do not have enough feature information, in this regard; iris recognition is dependent on the amount of information available in two iris images being compared. A typical iris recognition system commonly consists of four main modules as shown in Figure 1: • Acquisition the aim is to acquire a high quality image.