Noise Influence on the Fuzzy-Linguistic Partitioning of Iris Code Space
This work addresses the robustness of iris recognition systems for biometric security, but it is incremental as it applies existing fuzzy concepts to analyze noise sensitivity.
The paper investigates how noise affects the stability of fuzzy-linguistic partitioning of iris code space based on Doddington's biometric menagerie concepts, finding that the partitioning is unstable under noisy conditions, as demonstrated by tests on 180 iris recognition scenarios with added noise.
This paper analyses the set of iris codes stored or used in an iris recognition system as an f-granular space. The f-granulation is given by identifying in the iris code space the extensions of the fuzzy concepts wolves, goats, lambs and sheep (previously introduced by Doddington as 'animals' of the biometric menagerie) - which together form a partitioning of the iris code space. The main question here is how objective (stable / stationary) this partitioning is when the iris segments are subject to noisy acquisition. In order to prove that the f-granulation of iris code space with respect to the fuzzy concepts that define the biometric menagerie is unstable in noisy conditions (is sensitive to noise), three types of noise (localvar, motion blur, salt and pepper) have been alternatively added to the iris segments extracted from University of Bath Iris Image Database. The results of 180 exhaustive (all-to-all) iris recognition tests are presented and commented here.