Happy sad demarcation from facial images using neural networks

Abstract

The aim of this paper is to discuss the development of an efficient neural network that demarcates primary emotions of humans such as happiness and sadness. The neural network implements a back-propagation algorithm that minimizes the error in the network, by traversing back in the network and rectifying the weights of the nodes in the layers of the neural network. The artificial neural network takes inputs in the form of co-ordinate pairs that constitute the facial fiducial points around the lips. The three pairs of points are the six input layer nodes. These nodes are connected to two nodes in the hidden layer which are connected to two output nodes. The output layer nodes are the results of the network and these nodes are binary. If the emotion is happy, the first node will be active having a value 1 while second node will be inactive having a value 0 and vice versa if the emotion is sad. Using the data generated by the algorithm the artificial neural network is trained. The paper also compares the effect of normalization in neural networks by comparing the confusion matrix for normalized input data and raw input data. Keywords: Image analysis; Emotion Recognition; Neural Networks; Facial Images.