The traditional facial expression recognition method needs two independent steps of extracting features and classifying facial expression, and the deep neural network that attracted wide attention recently combines these two steps into one and performs well in the field of image recognition. We propose a method using partial Gabor features and neural network. First, we use the method of Ensemble of Regression Trees to identify facial significant landmarks and get important feature regions from them., such as the eyes, nose, and mouth; then extract the Gabor features using 24 Gabor kernel functions with 8 different orientations and 3 scales; finally classify them by the neural network of fully connected layer. As a result, we attained accuracies of 93.7% and 96.3% on JAFFE and CK+ expression databases respectively. The result shows that the combination of Gabor filtering and neural network can greatly improve the recognition accuracy.
Paper
Full text
Facial Expression Recognition Using Partial Gabor and Neural Network*
Semantic Scholar · Computer Science · 2018
Abstract
The traditional facial expression recognition method needs two independent steps of extracting features and classifying facial expression, and the deep neural network that attracted wide attention recently combines these two steps into one and performs well in the field of image recognition. We propose a method using partial Gabor features and neural network. First, we use the method of Ensemble of Regression Trees to identify facial significant landmarks and get important feature regions from them., such as the eyes, nose, and mouth; then extract the Gabor features using 24 Gabor kernel functions with 8 different orientations and 3 scales; finally classify them by the neural network of fully connected layer. As a result, we attained accuracies of 93.7% and 96.3% on JAFFE and CK+ expression databases respectively. The result shows that the combination of Gabor filtering and neural network can greatly improve the recognition accuracy.