Facial Expression Recognition based on The Fusion of CNN and SIFT Features

Abstract

Facial expression recognition is an important part of computer vision. In order to improve the accuracy of expression recognition, and at the same time overcome the problem that a single feature cannot fully represent the details of facial expressions, this paper proposes CNN and SIFT feature fusion algorithms: 1) using a custom CNN network, combined with the idea of the Inception module, that is, adding 1×1 convolution, can more efficiently use computing resources, extract more global facial expression information under the same amount of calculation; 2) use cascade regression to calibrate the facial facial structure points, and then extract SIFT features, so that the key points are concentrated on expression contributions In a large area, the two features merge with each other and complement each other. Finally, the fused features are classified using Softmax to improve the accuracy of facial expression recognition. Tested on the CK+, JAFFE and FER2013 data sets, the experimental results show that this method is an efficient method of facial expression recognition.