Emotion Estimation using Geometric Features from Human Lower Mouth Portion

Abstract

This paper presents approach for emotion estimation using geometrical features from the lower mouth portion. Basic geometric transformation features from lower face is extracted. A point based tracking method using Associative Recurrent Neural Network (ARNN) is developed whose input is most contributing features, identified from 65 dimensional data. The proposed approach effectiveness is realized on JAFFE and Yale data sets separately and collectively with good recognition rate for some basic emotions. © Springer International Publishing Switzerland 2015.