© 2017, IJARCSSE All Rights Reserved Page | 70 Volume 7, Issue 3, March 2017 ISSN: 2277 128X International Journal of Advanced Research in Computer Science and Software Engineering Research Paper Available online at: www.ijarcsse.com An Intelligent Approach for Anti-Spoofing in a Multimodal Biometric System P. Devakumar * M.Tech(Information Security), Dept. of CSE Pondicherry Engineering College, Puducherry, India DOI: 10.23956/ijarcsse/V7I3/0143 R. Sarala Assistant Professor, Dept. of CSE Pondicherry Engineering College, Puducherry, India Abstract— Biometric systems are vulnerable to certain type of attacks at various points in the biometric model. A spoofing attack which is submitting a stolen, copied biometric trait to the sensor to gain unauthorized access to the biometric system is one among them. Multimodal biometric systems are designed to increase the accuracy of the biometric system, but they are more vulnerable to spoofing attacks than a unimodal biometric system. The existing approaches for anti-spoofing do not consider multiple biometric traits and also have a high false acceptance rate. The proposed method is designed to overcome spoofing in a multimodal biometric system that uses a combination of face, fingerprint and iris images. The extracted biometric features are fused and fed to a convolution neural network that employs deep learning to detect spoofed features from real features. The proposed method gives better results than existing anti-spoofing methods. Keywords— Multimodal Biometrics, Anti-spoofing, Biometric feature extraction, Biometric feature fusion, Convolution Neural Network. I. INTRODUCTION The Multimodal biometric systems use more than one biometric trait. The Multimodal biometric systems use different mechanisms for biometric fusions. Multimodal biometrics are often referred to as multi- biometrics. Unimodal biometric systems often fail to correctly identify and verify an individual with a desired result and accuracy. Multimodal biometric systems are designed for this purpose. There are several kinds of attacks [1] in a Multimodal biometric system as shown in Figure 1 and they are type 1 attack include presenting fake biometrics at the sensor where a fake biometric sample is presented as input to the system. Type 2 attacks include replay attack where a biometric signal is stored and then it is replayed to access the system. Type 3 attacks include overriding the feature extraction where the feature extractor is attacked using a Trojan horse so that it produces feature sets selected by the intruder. Type 4 attacks include replacing features where the features extracted from the biometric input signal are replaced with a different feature set. Type 5 attacks include Corrupting the matcher where the matcher is attacked to produce preselected match scores. Type 6 attacks include tampering with stored templates where the attacker modifies a template in the Database. Type 7 attacks include attacking the channel between the stored templates and the matcher where the data sent to the matcher through a communication channel are modified. Type 8 attacks include overriding the final decision where the attacker is able to override the final match decision. A spoofing attack is a type 1 attack, where a stolen, copied biometric trait is submitted to the sensor to gain unauthorized access to the biometric system. It is used to defeat the biometric system. This kind of attack is also called as “direct attack” since it is carried out directly on the biometric sensor. The feasibility of a spoof attack is much higher than other types of attacks against biometric systems. Since it does not require any knowledge of the system. Figure 1. Types of attacks in biometric System