When biometric recognition is used for identification or verification it is important to assure the privacy of the data subject. This can be accomplished by using template protection mechanisms. These transform a feature vector that is derived from a data subject's biometric characteristic into a protected template (pseudo identity) and thus guarantee that no additional information such as health-related information is stored in the biometric reference. Due to noise, two biometric samples of one data subject are not the same and differ in some feature values (intra-class variations). This paper proposes a new template protection method which deals with these intra-class differences by applying cryptographic hash functions [10] in a step-wise manner to certain pieces of the biometric feature vector. This idea was inspired by Kornblum who proposed piecewise hashing for files in [7]. In this paper the method is applied to 3-dimensional facial data. The experimental results indicate that the biometric performance of the method is close to the biometric performance obtained without template protection.
