Facial Expression Recognition Using Convolutional Adaptive Neuro-Fuzzy Inference System (CANFIS)

Abstract

Facial Expression gives significant information about the emotion of a person. In the research area of computer vision, automated facial expression recognition is an important area because of its significance in Human Computer Interaction (HCI). To improve the level of interaction in man machine communication systems, extraction and validation of emotional information by facial expression analysis plays a major role. The proposed method in facial expression recognition is based on extracted features like mean, mode, minimum intensity, maximum intenstiy, Perimeter, Standard Deviation (S.D), orientation (about X and Y axis) and centroid (about X and Y axis). A random mask is generated to find the convolution of the feature set from the pooling layer to the fully connected layers of the Feed Forward (FF) Artificial Neural Network (ANN). A neuro-fuzzy based system called Adaptive Neuro Fuzzy Inference System (ANFIS) is used for classification by tuning the weight vectors in the fully connected layers of ANN which produces a robust algorithm called the Convolutional Adaptive Neuro-Fuzzy (CANFIS). The performance of this method is evaluated using Recognition Efficiency (RE) which yields an effective recognition system in comparison with the conventional ANFIS structure.

Keywords: Facial Expression Recognition, Convolutional ANFIS, Feature Extraction, Classification, ANFIS, Recognition Efficiency