Elham Reyhanian - School of Electrical and Computer EngineeringDr. Shariaty College, Iran
Sepideh Avizhe - School of Electrical and Computer EngineeringDr. Shariaty College, Iran
Khadije Sadra - School of Electrical and Computer EngineeringDr. Shariaty College, Iran

By the enhancement of technology, Biometric recognition has been used as an accepted method to authenticateusers, due to the fact that biometric data are unique for each individual. However, biometric data are highly privacysensitive and therefore they should not be revealed before and during processing. Basically, a biometric recognitionprocess constitute of two stages: feature extraction and classification; Among diverse algorithms of classification,neural networks are so popular, and exactly they are the most interesting point for attackers. In this proposal, two newprivacy preserving protocols are presented for Neural-Network-based Biometric Recognition (NNBR). In the firstprotocol, the general framework of previous works on privacy preserving neural network learning is used and modifiedin order to fit biometric recognition applications. Thus, inherited from mentioned works, our first protocol is a generalframework and it is run on discrete data. In the second protocol, the parameters and properties of neural network is usedto provide privacy for both biometric data owners and neural network owners and is carried out on continuous data,which makes it more consistent and precise than previous works in the sense that biometric data are continuous in theirnature and furthermore using continuous data remove errors produced during approximation. As a consequence,enjoying aforementioned features, the latter protocol is the first privacy preserving NNBR protocol among its kind.Finally, the use of image randomization in the preprocessing stage is suggested and discussed as a new method for dataprivacy and the trade-off between recognition rate and security is studied on a typical eigenface-based face recognitionsystem

privacy, biometric, face recognition, multi-layer perceptron neural network, discrete data, continuous data