MULTILABEL FACIAL EMOTION CLASSIFICATION

Abstract

A Multi label emotion recognition system is an important field of research and also has lots of applications. Examples include classify physically disabled people (deaf, dumb, and bedridden) and Autism children's emotional expressions based on facial landmarks. Understanding different human emotions is of great relevance as to gain insights into human cognition and affect as well as for the design of computational models and perceptual interfaces. Us as human beings expresses our feelings through distinct emotions yet most part, research studies have been limited to six basic categories—happiness, surprise, sadness, anger, fear, and disgust which restricts us from the real world problems. Here, in this paper we classified an important group of expressions, which we call compound emotion categories. Compound emotions are that can be visualized as combination of basic component categories to create new ones. For instance, happily surprised and angrily surprised are two distinct compound emotion categories. The dataset used in this project is CFEE_Database_230. The work is categorized in 3 major parts, first step is to analyze the training dataset and extract the region of interest (ROI) which is the most active region on the face when a person changes its facial expressions. Second step is to detect multiple image features so system can differentiate the facial landmarks in different emotional state of a person’s face. And finally in the third step the main approach for multi- label classification of the sets of emotions is to compare the difference in the value set of the relevant features for different emotion class and then labelling them as a category in the compound emotions. At last we analyse the result by applying the different classifier and observe that we get 92.2% accuracy rate using MULAN classifier.