Automatic Facial Expression Recognition Using Image Processing and Bayesian Regularized Recurrent Neural Network

Facial expressions can be considered as a means of communication by non-verbal signals. They are essential part of human relations. Automatic facial expressions recognition can be imperative for natural human-machine interaction. Automatic facial expressions recognition can be utilized in the field of behavioral science and in the health care department. Although humans perceive the facial expressions immediately and effortlessly, reliable automatic facial expression recognition by a machine is a challenging task. This paper proposes a hybrid method, comprising of image processing and Artificial Neural Networks (ANN) for automatic facial expression recognition. Image processing of the faces were carried out to extract the features. The facial features include the change in the shape and size of eyebrows, eyes and lips. The features extracted from the image are quantified and are used as training set for the artificial neural network. Two types of ANNs viz. feedforward neural network and Bayesian regularized recurrent neural network. Six types of expressions are recognized. The results are compared on the basis of confusion matrix, error histogram, mean absolute error, error plot and regression plot. On all the evaluation parameters Bayesian regularized recurrent neural network (BRRNN) is found to be best suited for automatic facial expression recognition.