Dynamic Eyes and Mouth Reinforced LBP Histogram Descriptors Based on Emotion Classification in Video Sequences

Abstract

In the world of visual technology, classifying emotions from the face image is a challenging task. In the recent surveys, they have focused on capturing the whole facial signatures. But the mouth and eyes are the utmost vital facial components involved in classifying the emotions. This paper proposes an innovative approach to emotion classification using dynamic eyes and mouth signatures with high performance in minimum time. Initially, each eye and mouth image is separated into non-intersecting regions from this video sequences. The regions are further separated into small intersecting sub-regions. Dynamic reinforced local binary pattern signatures are seized from the sub-region of eyes and mouth in the subsequent frames which shows the dynamic changes of eyes and mouth aspects, respectively. In each sub-region, the dynamic eyes and mouth signatures are normalized using Z-score which is further converted into binary form signatures with the help of threshold values. The binary signatures are obtained for each pixel in a region on eyes and mouth computing histogram signatures. Concatenate the histogram signature which is captured from all the regions in the eye and mouth into a single enhanced signature. The discriminative dynamic signatures are categorized into seven emotions utilizing multi-class AdaBoost categorizer algorithm.