English abstract
Human facial expression recognition (FER) is a challenging research problem due to the
feature extraction factor, which is an essential module for the FER system. In the case of facial
expressions, the feature extraction factor means getting the distinguishable feature for each
expression as a discrete symbol. Nowadays, modern artificial intelligent systems help to
emulate and gauge the reactions of the face. The detection and recognition with any machine
learning needs suitable training algorithms and then testing the data. This project aims to
identify the human emotions from an image and classify them into one of the existing classes.
The purpose of FER system is to implement the face detection algorithm presented by ViolaJones and try to detect all visible faces in every image. Then, the system crops the face and
recognizes the emotion displayed by comparing it with the stored classes in the dataset. This
research work employs a Histogram of Oriented Gradient (HOG) and Local Binary Pattern
(LBP) features to classify seven expressions: anger, fear, disgust, happiness, sadness, neutral
expression, and surprise. The system will determine the efficiency of the extracted features.
The feature selection process is performed by (ReliefF) algorithm. In the given thesis, facial
expression classification process is performed by Backpropagation Neural Network classifier.
Better recognition results are achieved with a hybrid system of HOG and LBP, where the
dimensionality is reduced up to thirty features instead of using thousands of features. This
reduction increases the recognition systems' speed and removes redundant information.
