Biometric discretization based on reliability-dependent bit allocation and linearly separable subcode encoding

Biometric discretization converts biometric features into a binary string via a feature space quantization and a subsequent encoding process. The classification accuracy of the discretized features can be improved by weighting the features according to their discrimination power through a bit allocation process in such a way that more bits are allocated to more discriminative features and vice versa. This paper presents a discretization scheme based on reliability-dependent bit allocation and Linearly Separable Subcode encoding to extract discriminative binary strings for biometric representation. The superiority of our approach is experimentally vindicated using two popular face data sets.