Face Recognition Research Based On The Traditional Algorithm And The Convolutional Of The Neural Network

In daily life,face recognition technology has brought increasing convenience to people,such as taking high-speed trains,paying with a face swipe,and punching in at work.In addition,video surveillance,information security and criminal investigation are also its important application fields.Traditional face recognition methods are good at extracting shallow face features,which lack of good discrimination.Therefore,they can not be more effective.The rapid development of deep learning and the continuous emergence of various types of face datasets promote the further development of face recognition technology.As a result,face recognition based on deep learning model can achieve higher accuracy.In practical application scenarios,there are many factors such as lighting,posture,etc.will affect the recognition effect to a certain extent.This paper combines the traditional algorithm with the deep learning technology and optimizes the Convolutional Neural Network(CNN)structure,then add the channel attention mechanism into it to improve the effect of face recognition.1.A face recognition method combining Gabor wavelet and lightweight convolutional neural network is proposed.There are many deep convolutional neural network models,and they have achieved high accuracy in face recognition task,but they have a large amount of calculation,high resource consumption,and there will be the loss of local features of the face.Lightweight convolutional neural networks have fewer parameters and lower computational complexity.Gabor wavelet is sensitive to the edge of the image and is not easily affected by lighting and other factors,so it has better effect in extracting local texture features of the image.In order to effectively play the advantages of both,this paper proposes a method of combining Gabor wavelet and lightweight convolutional neural network.Firstly,the functions realized by Gabor filter are integrated into a convolution layer(Gabor layer),and then the GMNet network is combined with Gabor layer and lightweight convolution neural network to extract face features and finally complete the face recognition task.The experimental results show that the improved model can improve the performance of face recognition and effectively resist the change of illumination,posture and age.2.A face recognition method based on improved residual network and channel attention mechanism is proposed.CNN performs very well in face recognition.However,deep convolutional neural networks have difficulty in convergence and optimization during the training process.The emergence of residual networks alleviates these problems.In addition,the channel attention mechanism can help networks to selectively learn features that contain more useful information,which can enhance the expressive ability of the network.Therefore,we first use the Mish function to optimize the gradient descent effect,and the improved residual network RESNET＿IR is obtained,then the Channel Attention Module(CAM)is introduced into it to obtain the final network model named CAMRESNET＿IR,making the extracted face features more discriminative.The experimental results show that our model can improve the accuracy of face recognition,and can maintain better results when the illumination,posture,and age change.