| Face recognition technology as one of the important branches of the field of machine vision,in the security monitoring,attendance card,identity authentication has been widely used.With the development of deep learning,face recognition technology has made great progress in recognition efficiency and accuracy.However,face recognition technology is prone to be affected by lighting,Angle,posture,occlusion and other actual application environment.The complex environmental background of ATM placement in the bank is easy to interfere with the accuracy of face recognition.Therefore,further research on face recognition technology is needed.This paper studies the improved Retinaface face detection algorithm and the improved VGGnet face recognition algorithm,and wraps the designed algorithm in the ATM identity authentication system.The specific research is as follows:In the aspect of face detection,an improved Retinaface face detection algorithm is designed.Face detection as the first step of face recognition,the accuracy of face detection determines the accuracy of face recognition.Firstly,four data enhancement methods,namely light distortion,geometric distortion,random erasurement and random Mosaic,were used to enhance the data of 6,000 face images with complex background selected from the Widerface dataset to 36,000,making the new data set more in line with the complex detection environment of the bank.Secondly,the Distance-Io U loss function is used as the regression loss function of the anchor frame in the model to make the optimization objective consistent with the regression evaluation index and improve the detection accuracy of the anchor frame.Finally,five inception structures are embedded in the Retinaface model to deepen the depth of the network layer,extract richer features,and improve the detection rate of small-scale faces.The results show that the improved Retinaface model has better performance,the recall rate and precision rate are both higher than the original Retinaface model,and it can meet the requirements of real-time detection in cash machines.In the aspect of face recognition,an improved VGGNET face recognition algorithm is designed.The VGGNET model with superior performance and high precision is used as the basic model of the network.Considering the speed requirement of real-time identification,the paper firstly reduces the parameters of VGGNET 16 network.By replacing the first two layers of the last three full connection layers with the pooling layer,the network parameters are reduced by about 60 percent,the accuracy remains unchanged,and the speed is about half of the original.Secondly,channel attention mechanism is embedded into the VGGNET-16 network model,so that the weight of each channel can be automatically recalibrated to improve the weight score of small faces.The Arc Face loss function,which is more in line with face recognition,is used to make the distance between the same person's face features as small as possible,and the distance between different faces as large as possible,and strictly distinguish from the background,so as to improve the accuracy of face recognition.Finally,through the experimental comparison of different improvement strategies and different models,the experimental results show that the recognition accuracy is above 0.95,and the designed face recognition model can accurately recognize faces.Finally,the face detection algorithm and face recognition algorithm are packaged into the bank ATM system.First of all,the design of the system scheme,the bank ATM identity automatic authentication system environment,the overall structure,each module corresponding to the system process,system interface was introduced in detail.Then the design of the bank ATM identity automatic authentication system test.The results show that the designed automatic identification system of ATM based on face recognition has high accuracy and usability,and can be used in real-time environment for face recognition,so as to authenticate the identity. |
PDF Full Text Request