Download now Free registration required
This work presents a preliminary statistical analysis on the reliability of soft biometrics systems which employ multiple traits for human identification. The analysis places emphasis on the setting where identification errors occur mainly due to cross subject interference, i.e., due to the event that subjects share similar facial and body characteristics. Finally asymptotic analysis provides bounds which insightfully interpret this statistical behavior. Soft biometrics, as defined by Jain, are those characteristics which provide weak biometrical information about an individual, but lack distinctiveness and permanence, and thus do not suffice to fully identify a person. This lack of distinctiveness can be partially overcome by employing multiple traits to classify individuals in pre-defined categories.
- Format: PDF
- Size: 140.6 KB