Research And Implementation Of Large Scale Face Recognition Technology Based On Improved DeepID

As one of the biological characteristics of the human body,the human face has a strong distinction as same as fingerprints and voices,and its unique uniqueness and non-reproducibility provide the necessary prerequisites for identity authentication.Predecessors have tried a lot of methods to solve the face recognition in the natural environment.Early research is usually better by using a shallow model that uses a complete low-level feature description.In recent years,with the improvement of hardware performance,deep learning with its unique self-adaptability,high flexibility,strong goodness of fit,and gradually gains amazing figures in various fields.It can learn from a variety of samples of different types of values to get the functions we need.This powerful ability to approximate the objective function makes it the most effective way to solve certain types of problems,such as learning the data(images,sounds,radar reflection signals,etc.)obtained by the sensors.Years ago people began to try to apply the neural network method to face recognition.It can adapt to a variety of scenarios and has a strong practicality.Convolution neural networks(CNN)have been proven to be effective in extracting advanced visual features.There are internal relationships between the various features in the face image,and this characteristic can be represented by the correlation between the units in the artificial neural network.In the field of face recognition,with the most outstanding performance,Deep ID provides a way to learn advanced features and using convolution neural network to build deep models.In this paper,the general rules that exist in the Deep ID network has been found by analyze the data of the three generations of Deep ID models;the effects of various parameters in the network on the accuracy of classification have been verified through experiments;eventually,through scientific structural simplification and improvement to three generations of the Deep ID network.This method finds the way to improve the performance and accuracy of the network,reducing the dependence on data set.Finally the amount of parameters of the network has been greatly reduced and the running speed of the model has also been more significantly increased.