Paper: Binary Dense Convolution Networks for Expression Recognition 

26 Oct 2018 in Blog on Paper

Abstract: The research of facial expression recognition has become an important topic in the field of artificial intelligence. However, the requirement for huge computing resources has limited the application of traditional convolutional neural networks. Since the binary neural network replaces the floating point multiplication arithmetic by fast AND OR arithmetic, the need of computing resources can be greatly reduced. In this paper, we propose a facial expression recognition algorithm based on data enhancement and binary convolutional neural network, and 66.15% expression recognition accuracy is obtained on the dataset FER2013. The algorithm has surpassed some convolutional neural network algorithms based on floating point multiplication arithmetic, which makes it possible to transplant expression recognition algorithms into small devices.

Key words: deep learning; data enhancement; binarization; dense convolutional neural network; expression recognition

NOTE: This paper is compeleted by G.Z.Wen, Y.H.Ma, and I, which is one of our works of the National Research Trainning Project. If you are interested in the paper on Binary Dense Convolution Networks, you can read the full Chinese pdf version. It will be published in the journal “Computer and Digital Engineering” in 2020.