Hyperspectral Face Recognition Based On Deep Learning

With the development of remote sensing technology,spectral imaging has been widely applied in agriculture,biomedicine and other aspects.Researchers have shown that different parts of the face have huge spectral variability,it is the spectral variability captured by the hyperspectral imaging that hyperspectral images possess more discriminative information.Therefore,hyperspectral imaging technology can be used for face recognition.At present,the high spectral face recognition technology is mainly based on the traditional algorithms such as Gabor,HOG,LBP and so on.With the rapid development of artificial intelligence,the deep learning technology have made great progress in speech recognition,face recognition,automatic driving,and all various medical treatment.Supported by the big data,deep learning method has more advantages than the traditional algorithm,this is because that deep learning neural networks can extract the distinctive features from high dimensional and complex information compared with the traditional algorithm.Hyperspectral images have more bands,higher dimension and more information than RGB images.Therefore,it is of great significance to apply the deep learning method to the hyperspectral face recognition.The deep convolution neural networks,as a deep learning method,use local receptive field and shared weights to recognize images which have quality of shift,scale,and distortion invariance.In this paper,we studied the traditional algorithms and deep convolution neural network based hyperspectral face recognition.The main work of this paper is as follows:Firstly,for the traditional algorithm,a covariance based image fusion method was proposed,which mapped multi-dimensional hyperspectral face images into twodimensional gray images,and then extracted the LBP features of two-dimensional images.In the end,the chi square distance was used as classifier.Secondly,for deep learning method,a algorithm of hybrid channel was used to achieve data augment;training a 12 layer convolution neural network model for hyperspectral face recognition and comparing with different models;according to the high spectral facial data with the multi spectral characteristics,we proposed a multi band recirculating training network and experiments have shown the effectiveness of our proposed method.In the opening datasets CMU-HSFD and PolyU-HSFD,we got the highest recognition rate with the optimized manual feature extraction algorithm.With the deep learning method,our network model also achieved the highest recognition rates than other deep learning model under a limited datasets.