Person independent analysis of facial emotion

Abstract

There are billions of billion communicating messages that are transmitted and received every day. This huge amount of messages is used in various types of communication from traditional form like oral communication to modern technology such as texting and emails. However, in terms of effective and efficient communication, emotions play an important role in conveying messages. With this powerful and natural tool, human is able to express their feelings and intensions. Although it is very easy for human to recognize the emotions, computers or machines are not able to do it with such accuracy. Facial emotion still remains a challenging area to engineers.

Since the human’s emotional expressions are complex and subtle, it is difficult for computers to determine all possible emotions. As a result, there are 6 basic emotions that are chosen to use in the project: Happiness, Sadness, Anger, Fear, Surprise and Disgust. In this project, the goal is to achieve effective recognition of these emotions by extracting facial features and classifying them. In order to extract facial features, Histogram of Orientation Gradients (HOG) and LBP (Uniform LBP and RIU-LBP) are chosen. These methods will be combined under feature fusion. HOG descriptors are chosen because they significantly outperform existing feature sets for human detection. LBP is computational very simple and can be useful in representing textures. Classifier such as SVM is use for classification of the features.

The robustness of the aforementioned system is tested against Cohn-Kanade database, which contains images of 6 basic emotions.