Multiple Face Image Feature Extraction Using Geometric Moment Invariants Method

摘要

Research on human facial expression recognition has become a growing field. One important step in the recognition of facial expressions is feature extraction. This research uses Geometric Moment Invariants (GMI) as a feature extraction method. Research on facial expression recognition using either the GMI method or another method use single face image as the dataset. Therefore, in this study uses GMI feature extraction to classify facial expressions on multiple face images. Face detection process uses Viola-Jones method on OpenCV and classification process uses Multi Class SVM method. The results are features for each expression and a small average accuracy of 5 times. Therefore, the classification is also done with the k-fold cross validation technique with another classification method. The average accuracy results are still small. It caused by the training image also using outer area of face in the image, so the background included as the image features. It is tested from k value 2 to10, and produce Multi Class SVM 10.2%, Decision Tree Classifier 14.73%, Random Forest Classifier 14.78%, Gaussian Naive Bayes 14.73%, Nearest Centroid 14.66%, MLP Classifier 11.09%, and Stochastic Gradient Descent Classifier 14.19%. The highest accuracy result is Random Forest Classifier method 14.78%. In Random Forest method, the best k value obtained is 4 with an average accuracy 16.18%.