Face Recognition Based On Broad Learning Network

The human face is a relatively stable biological feature,and it is also a relatively secure authentication method.Face recognition has attracted many scholars' interests because of its broad application prospects.Researchers have proposed many recognition algorithms,some of which have gradually matured,and have been practically applied in the fields of security,finance,and driverless.In recent years,face recognition based on deep learning has become a hot spot and has achieved a high recognition rate.However,the face recognition algorithm model based on deep learning takes a long time to train,has high computational complexity and consumes a lot of computing resources.Aiming at this problem,this paper studies the face recognition problem based on the new machine learning method of width learning.The main work is as follows:First,a face recognition algorithm based on broad learning is proposed.The algorithm first uses gray histogram equalization to eliminate the influence of lighting,and then uses Fisherfac to extract face features.The obtained features are used as the input of the broad learning network to train the broad model.Since the broad learning network is a single hidden layer network,the training time is greatly shortened by using the least squares algo-rithm for training.In addition,directly inputting the Fisher features of the face,the feature enhancement layer of the broad network further enhances the Fisher features,obtaining the high-dimension features of the face.Experimental results show that the algorithm achieves high recognition accuracy while occupying less computing resources.Secondly,an expression recognition algorithm based on broad learning is proposed.Different from other algorithms in the literature,this algorithm directly uses low-resolution small-sized expression images as input to train the broad learning network without the fea-ture extraction.The proposed algorithm uses the feature enhancement layer of the broad learning network to extract the potential features of the input images.In the experiment,the proposed algorithm was tested in the JAFFE database,and the experiment results showed that the recognition rate of the proposed algorithm was higher than the methods in the ex-isting literature.Finally,a face recognition algorithm combining broad learning and extreme learning machines is proposed.The algorithm borrows from the idea of auto-encoder,and first inputs the face images to the broad learning network,and learns auto-encoder,which means that the output of the broad network is equal to the input face images.The weight obtained by learning the broad auto-encoder as a face feature.These weight features are further used to train the extreme learning machine to complete the face recognition process.The algorithm is tested with the ORL face database,and the results show that the algorithm can obtain a high recognition rate in the presence of uneven illumination and incorrect posture on the face.