Quantifying Quality: A case study in fingerprints
Sharathchandra Pankanti, Norman Haas, Nalini K. Ratha, Rudolf M. Bolle
presented Tarrytown, NY, March 2002, Proc of IEEE Conference on AutoID 2002.
For a particular biometric to be effective, it should be universal: every individual in the target population possesses the biometrics and every acquisition from each individual provides useful information for personal identity verification or recognition. In other words, everybody should have the biometrics and it should be easy to sample or acquire. In practice, adverse signal acquisition conditions and inconsistent presentations of the signal often results in unusable or nearly unusable biometrics signals (biometrics samples). This is confounded by the problem that the underlying individual biometrics signal can vary over time due, for example, to aging. Hence, poor quality of the actual machine sample of a biometrics constitutes the single most significant cause of poor accuracy performance of a biometrics system. Therefore, it is important to quantify the quality of the signal for either seeking a better re-presentation of the signal or for subjecting the poor signal to alternative methods of processing (e.g., enhancement). In this paper, we explore a definition of quality of fingerprint impressions and present detailed algorithms to measure image quality. The proposed quality measure has been developed with the use of human annotated images and tested on a large number of fingerprints of different modes of fingerprint acquisition methods.