The increasingly rapid convergence between machine and human languages has created a growing interest in the semantic analysis of multimedia contents. Despite the recent technological progresses, much remains to study in order to fill the semantic gap, the lack of coincidence between the information the machine extracts from raw data and the way humans interpret that content. Soft Biometrics (SB) represent the recent advances that Biometric research performed in these regards.
Those traits have attracted the attention of the research community in that they have some characteristics peculiar to Hard Biometrics while they improve the semantic content they carry. In this dissertation we explore the Body SB concept both from a theoretical and a practical point of view. Taking advantage of dataset used for medicals and demographics studies, we analyze the relation between body parts and personal traits like anthropometric measures, weight, and gender.
The ability to extract this information is tested by using both the 2D image analysis and by processing 3D video streams. We use SB for pruning a biometric database, enabling faster and more accurate response of a face recognition system. Moreover, we build an application that demonstrates the feasibility of people re-identification for limited groups of users. Furthermore, we present two medical applications for health conditions tracking. The former enables the interaction between the user and an automatic system that performs a medical check up of user's body, providing hints and lifestyle suggestions. The latter is intended to support the monitoring of cosmonauts' weight losses due to the lack of gravity in outer space.
