ABSTRACT
Robust and efficient emotion recognition is very important and challenging for many physiological and augmented applications. These applications however, usually suffer from the non-emotion-related varieties, the "individual varieties", especially when cope with multiple subjects. In this paper, we proposed and presented an improved grouping structure called the Self-Adaptive Biometric Signatures Based (SABSB) emotion recognition system. The system is unique in that it transforms a traditional subject-independent problem into several subjectdependent cases to remove the influence of individual varieties. Also, the selfadaptive procedure allows a flexible model generating process which is more practical for real applications. The proposed framework is implemented and tested using mixture multivariate t-distributions (mmTD). Results are compared with conventional emotion recognition system as well as the original BSB system.
