A Noval Distributed Architecture for Expression Recognition

Abstract

A distributed facial expression recognition approach based on MB-LGBP feature and decision fusion is presented in this paper to accomplish subject-independent facial expression recognition more efficiently. At first, the Multi-scale Block Local Gabor Binary Patterns (MB-LGBP) are extracted from expression regions to achieve both locally and globally informative features. Then a distributed architecture is proposed to accelerate the recognition process, in which features of each single region are utilized to perform expression classification in parallel. The final decision is made by an artificial neuron network (ANN) based data fusion of the confidence information got from the classification of each region. In experiment, we compare the runtime and recognition accuracy of our system with several other popular expression recognition paradigms. The results show that the distributed architecture can promote the efficiency of facial expression recognition prominently with comparative performance in recognition accuracy.