Facial Expression Recognition Based on MB-LGBP Feature and Multi-level Classification

Abstract

This paper presents a facial expression recognition approach based on MB-LGBP feature and multi-level classification. First, the multi-scale block local Gabor binary patterns (MB-LGBP) operator is extracted to achieve both locally and globally informative features. Then a two-level classification method is proposed. At the coarse level, two expression candidates with the first two high decision confidence are selected from 7 basic expression classes based on MB-LGBP features. At the fine level, one of the two candidate classes is verified as final expression class based on more delicate 2D MB-LGBP features. The promising result proves the superiority of our method to some other popular paradigms in expression recognition.