Pose-varied Face Recognition Based On Hadoop

With the development of society and technology,face recognition is widely used in the information security,financial security,anti-terrorism,criminal investigation and other fields because of its characteristics such as intuitive results,well hidden and convenient operation.But in the unconstrained environment,the cameras usually can't collect the effective faces,such as shaded face,complex environmental interference and the change of expression and posture,and it will lead to a certain degree of decline in recognition rates.In addition,with the continuous expansion of the data generating points,image data is growing at an annual rate of 20%.Therefore,the cloud computing technology is adopted in the traditional identification methods to find the target face image quickly in large amount of data recently.Firstly,this paper has built a cloud platform based on the Hadoop structure,and then the pose-varied face recognition under the unconstrained environment research work has been done base on the Hadoop cloud computing platform.The following work has been done in this paper:Analyzing the current popular Hadoop cloud platform structure,and using VMware workstation as virtual machine,establishing three ubuntu system and composition structure of Master/Slaver,building the Hadoop distributed system by configure the related network,file and software.Analyzing the composition and characteristics of traditional neural networks,and the structure of convolutional neural network,doing further research work on the convolutional neural network's convolution layer,pooling layer and active layer,and summarizing the characteristics of convolution neural network finally.Through the experiments on the current popular Le Net-5 convolutional neural network model,some research work about analyzing the critical factors which has a important influence on the CNN,such as the scale of slide-window,the depth of the constructional and the pooling method has been done based on CAS-PEAL face database.In order to enhance the generalization ability of the model,we make a further comparison on various nonlinear transformation technology such as activation function and regularization method.This paper has used the Hadoop cloud platform to collect and arrange the image sets,then using the improved convolutional neural network to extract thefeatures of the collected pictures,and using the PCA algorithm to reduce the extraction features,finally,the cosine similarity measure is used to recognize the target face in the recognition phase.The proposed pose-varied face recognition algorithm has been tested effectively on the CAS-PEAl face database.