Human Emotion Recognition Using Gabor Variance Features with Back Propagation Neural Network Classifier

It is natural as stimuli-response humans always express their feelings & reaction to any certain event automatically appears on the 'face,. Facial expression is an important medium for human communication as it express human thinking, feelings and his or her current mental situation, thus it is being used in many application areas. This paper aims to introduce a novel method for human emotion recognition using average variance as the feature vectors obtained from the Gabor filter convolved 'n, images which helps in classifying those emotions. Based on the Gabor variance features, a three layer back propagation neural network (BPNN) has been used as a classifier. The BPNN architecture used in the experimentation work contains 210 input units in the input layer, which corresponds to the displacement information of the Gabor variance feature vectors. There are 6 units in the output layer and one hidden layer of 256 units. According to the JAFFE database, the average accuracy of the proposed emotion recognition algorithm was the highest with 94.66%. Timing analysis using the same database shows that the template response time is lower because the BPNN is only 3-tier architecture which requires a single training as emphasis is on recall time. Because network training is only needed once, recall time is more important than training time.