FACIAL EXPRESSION RECOGNITION FOR THAI SIGN LANGUAGE IMAGE

This thesis presents an algorithm for facial expression recognition to help the interpretation of the meaning of Thai sign language image by adding sentimental information. Facial expressions are the facial changes in response to the signer’s internal emotional states. Some facial changes are similar. This similarity may cause the reduction in the facial expression recognition rate. We thus propose facial expression recognition framework in this thesis. The framework consists of the three components: face detection, facial features extraction and facial expression recognition. The face region is first automatically detected from input images by using Viola-Jones algorithm. The face components such as eyes and mouth are then extracted and normalized. Then, these features are combined to form feature vectors that will be used in the recognition. In the recognition step, K-Nearest Neighbors (KNN) and Support Vector Machine (SVM) are used to classify the facial expressions. In our algorithm, there are three datasets: positive, negative and neutral. For the dataset, we select the images that have the dissimilar facial changes as the input images. Positive dataset based on the positive expression that is identified by the facial changes: raising of the eyebrows and pulling of the mouth corner. For negative dataset, lowering of the eyebrows and puckering of the mouth are chosen to identify negative expression. For neutral dataset, expressionless facial changes are selected as the neutral expression. In this paper, we compare the classification accuracy by recognizing the single facial change and the combination of the facial changes for facial expression recognition. The experimental results show that the classification accuracy of the mouth is better than that of other parts of the face.