Antitza Dantcheva - Doctorant - PhD Student Multimedia Communications
Date: - Location: Eurecom
This work presents the concept of soft biometric systems (SBSs), as systems that employ weak semantic human traits towards human (re-)identification, database search pruning and facial aesthetics prediction. Such traits include gender, age, presence of glasses, hair and skin color. SBSs inherit the non-intrusiveness and computational efficiency, which allow for fast, enrolment-free and pose-invariant biometric analysis, even in the absence of consent and cooperation of the surveillance subject. Human (re-)identification Hereby we shed some light on the statistical properties of pertinent parameters to the proposed system, like employed traits and traits-instances, all over categories, size of an authentication group, spread of effective categories and correlation between traits. Further we introduce and elaborate the event of interference (a subject picked for authentication is indistinguishable from another subject in the same authentication group). This second application employs soft biometric traits to pre-filter a given human image database, with other words to prune the search. We explore the case where such pre-filtering is prone to error, in which case the filtering-gain comes at the risk of missing the target of the search. Finally we connect soft biometrics to subjective evaluation of female facial aesthetics. This approach considered both permanent, as well as non permanent facial characteristics and expression. The study explored the role of a specific set of features in affecting the way humans perceive facial images. Based on this study we construct a metric for female facial aesthetics prediction.