The Research of Chinese Ethnical Face Recognition Based on Deep Learning

摘要

Face recognition emerged in the seventies. With the introduction of deep learning methods, especially the convolution neural networks (CNNs), more and more traditional machine learning techniques have been recently superseded by them. In a multi-ethnic country like China, the study for Chinese ethnical face recognition (CEFR) has practical demands and applications. In this paper, we provide a brief of popular face recognition procedure based on deep learning method firstly. Then, as lacking of the corresponding dataset, we construct a collection of Chinese ethnical face images (CCEFI) including Han, Uygur, Tibetan and Mongolian. Based on multi-task cascaded convolution networks (MTCNN) and residual networks (ResNets), our proposed model can achieve promising results for face detection and classification. Specifically, the average precision reaches 75% on CCEFI self-draft. Experimental results indicate that our model is able to detect the face in some constrained environments and distinguish its ethnical category. Meanwhile, the dataset established by us would be a useful dataset for relevant work.