An Efficient Facial Feature Extraction Method Based Supervised Classification Model for Human Facial Emotion Identification

摘要

Certain situations require human intervention to identify the emotional condition of the subject. In order to automate this process, the paper explores the idea of applying the Fisher face method to build an accurate, reliable and automatic system accompanied by supervised classification methods such as K-nearest neighbor and artificial neural network for human emotion recognition as depicted in still images. The experimental analysis is performed on the popular Japanese female facial expression (JAFFE) database. We have explored the efficacy of the Fisher face model in classifying human emotion based on content obtained from facial images. We have used basic prototypic emotions as defined by Paul Ekman - happiness (H), sadness (S), surprise (SU), and neutral (N), as our four classes. The technique used for feature extraction of facial expression detection is local fisher discriminant analysis (LFDA) which performs dimensionality reduction. The supervised classification task is carried out by K-nearest neighbor (KNN) and artificial neural network (ANN) separately. It is observed that ANN performance is better in case of emotions like happiness and surprise while neutral and sad emotions are recognized better by KNN technique.