Impact of trivial quantisation on discrimination power in biometric systems

Trivial quantisation is widely used in cancellable biometrics and bio-cryptosystems for error tolerance. This method segments the feature domain into several non-overlapping intervals of equal size, whereas all features in the same interval will be considered matched. Although it is intuitive that trivial quantisation brings about the boundary issue, similar features near interval boundaries may be mapped into different intervals and lead to matching failure. Previous works do not provide specific theoretical analysis on how trivial quantisation impacts discrimination power (DP), an important performance index in biometric systems. Assuming genuine features and imposter features follow Gaussian distributions, the DP provided by the conventional matching method and trivial quantisation-based matching method from a theoretical perspective is discussed and compared. Also, formulas are developed for calculating the error tolerance parameter that maximises the DP in each method and the discrimination loss caused by trivial quantisation.