Investigating fusion approaches in multi-biometric cancellable recognition Anne M.P. Canuto ⇑ , Fernando Pintro, João C. Xavier-Junior Department of Informatics and Applied Mathematics (DIMAp) Federal University of RN, Brazil article info Keywords: Biometric template protection Multibiometric recognition abstract Cancellable biometrics has recently been introduced in order to overcome some privacy issues about the management of biometric data, aiming to transform a biometric trait into a new but revocable represen- tation for enrolment and identification (verification). Therefore, a new representation of original biomet- ric data can be generated in case of being compromised. Additionally, the use multi-biometric systems are increasingly being deployed in various biometric-based applications since the limitations imposed by a single biometric model can be overcome by these multi-biometric recognition systems. In this paper, we specifically investigate the performance of different fusion approaches in the context of multi- biometrics cancellable recognition. In this investigation, we adjust the ensemble structure to be used for a biometric system and we use as examples two different biometric modalities (voice and iris data) in a multi-biometrics context, adapting three cancellable transformations for each biometric modality. Ó 2012 Elsevier Ltd. All rights reserved. 1. Introduction The development of safe techniques in authentication systems is an important requirement in different fields of our modern inter- connected society. The constant improvement of communication technologies and new media platforms has caused an increasing capability of interaction among individuals. However, as interac- tion and access to information increase, there seems to have the same proportion of fraudulent attacks, making the development of efficient security techniques an important issue (Jain, Nandakumar, & Nagar, 2008). The fact of human biometrics are unique to each individual, cannot be lost, forgotten or stolen, has led to an increasing interest in biometric-based systems in order to increase reliability, convenience and universality of authentication systems (Bolle, Connell, & Ratha, 2002; Maltoni, Maio, Jain, & Prabhakar, 2009). An important concern in biometric-based systems is related to the security of biometric data, mainly when a biometric template is compromised, either through lost, stolen or unauthorized copies. Biometric information is permanently related to a specific user and they can not be changed if they are compromised. In this sense, the development of techniques that allow the use of biometric data without compromising the system security is fundamental. The idea of cancellable biometrics has been proposed to address such security issues (Jain et al., 2008). Cancellable biometrics consist of applying functions (generally not invertible) in the original bio- metric data in order to obtain transformed or intentionally- distorted biometric data. Once the biometric data is transformed, an authentication system will use this transformed data for pro- cessing and storage. Therefore, in case of being compromised, a biometric template can be replaced through the application of a new transformation function (Maltoni et al., 2009). Although the use of cancellable biometrics bridges the gap between the convenience of biometric authentication and security vulnerabilities, the use of such transformed data generally de- creases the performance of the biometric-based system. This is be- cause the level of complexity for the transformed biometric can be much higher than for the original data. For that reason, cancellable biometric systems must support a good trade-off between discrim- ination capability and non-invertibility (high security), when applied to any biometric modality. One important aspect to improve the recognition performance of biometric-based systems is the fusion of multiple sources of bio- metric information, through a multi-biometric recognition system. It aims to combine the identity evidence presented by multiple sources of biometric information in order to enhance the classifica- tion performance (Ross, Nandakumar, & Jain, 2006). When such systems are used, more discriminating information available per individual is used and it tends to improve the performance of bio- metric systems. Multi-biometric systems can offer some advanta- ges over single biometric systems, such as: (1) To improve significantly the accuracy of the biometric authentication (identifi- cation or verification) process; (2) To provide a high degree of flex- ibility, since unusable or non-preferred biometric traits in particular individuals can be compensated by information of other biometric modalities; and (3) To provide an additional difficulty to avoid spoofing attacks. As a multi-biometric recognition system is expected to improve accuracy and defeat spoofing attacks, template protection methods 0957-4174/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.eswa.2012.10.002 ⇑ Corresponding author. Tel.: +55 84 32153814x221. E-mail addresses: anne@dimap.ufrn.br (A.M.P. Canuto), fernando.pintro@ hotmail.com (F. Pintro), jcxavier01@gmail.com (J.C. Xavier-Junior). Expert Systems with Applications 40 (2013) 1971–1980 Contents lists available at SciVerse ScienceDirect Expert Systems with Applications journal homepage: www.elsevier.com/locate/eswa