Soft Biometrics; Human Identification Using Comparative Descriptions.

Area of Science:

• Biometrics and Human-Computer Interaction
• Pattern Recognition and Machine Learning

Background:

• Soft biometrics utilize human-describable traits for identification, aiming to bridge the semantic gap between biometric data and verbal descriptions.
• Traditional soft biometric methods often rely on absolute labels, which can be unreliable and lack discriminatory power.

Purpose of the Study:

• To introduce a novel method for obtaining soft biometric descriptions using comparative categorical labels.
• To enhance the objectivity and discriminatory capabilities of human-derived biometric descriptions.
• To demonstrate the robustness and accuracy of soft biometric signatures derived from comparative descriptions.

Main Methods:

• Developed a novel method using comparative categorical labels to describe differences between subjects.
• Applied the Elo rating system to infer relative measurements from comparative human descriptions.
• Utilized a support vector machine to derive relative measurements from gait biometric signatures.

Main Results:

• Comparative descriptions provide more objective information and increased discriminatory capabilities compared to absolute labels.
• Soft biometric signatures derived from comparative descriptions are robust and enable accurate subject recognition.
• Relative measurements can be successfully obtained from gait biometrics using human comparisons.

Conclusions:

• Comparative descriptions offer a more reliable approach to soft biometric identification.
• The proposed method effectively bridges the semantic gap in biometrics.
• This technique allows for subject retrieval from video footage based on human comparisons.