Research And Implementation Of Face Recognition Method In Sign-in System

In recent years,with the enhancement of computer computing ability,artificial intelligence technology has been developed at a high speed,especially in the direction of computer vision.Face recognition plays an important role in computer vision and has been widely used,but fast and accurate face recognition is not a very easy thing.In recent years,face recognition system has been applied to all walks of life,people's demand for face recognition system is becoming more and more intense.Face-based check-in system,because of its convenience,intelligence,high efficiency and other advantages,is used for meeting check-in,event check-in and classroom check-in and other scenarios.At present,face check-in system face the challenge of uncertainty such as lighting,masking and posture.In the current face recognition algorithm,the traditional extraction method based on local features can deal with the situation of low light intensity and small influence of light,but in natural light,the impact of complex light cannot be overcome.The face recognition method based on convolutional neural network can extract the deep abstract features of the face,but the characteristic differentiation of the extraction is not enough for the problem of masking and attitude.To this end,this paper on the face recognition technology to carry out research as follows:First,for a single local feature cannot effectively deal with face recognition under natural light,a multi-feature fusion method is proposed to extract facial features,using broad discrimination analysis for nonlinear dimension reduction,can also reduce the impact of light when reducing parameters.Using differential correlation to blend features can eliminate interclass correlation and maximize the pair correlation between the two feature sets.In this paper,the multi-feature fusion method proposed in this paper can effectively extract features from a variety of lighting conditions under natural light,and achieve better results than a single local feature extraction method.Second,in view of the convolutional neural network using Softmax Loss,which cannot extract distinguished features,a face recognition method based on DenseNet is proposed,which uses L2-Softmax loss and Island loss joint training,which not only extracts features with compact interclass differences and separable interclass differences,but also reduces the parameter amount of the model.The comparison experiment verifies the validity of this method.Third,face check-in system by lighting,masking and other influences,in response to this problem,the system will be multi-feature fusion method and convolution-based neural network method combined,the use of C++ as the main development language,Visual Code Studio as a development tool,design and implementation of the classroom check-in system,the system mainly includes face detection,face recognition and server three modules.The system basically meets the functional and performance requirements of daily use.