Research And Application Of Video Face Recognition Based On Deep Learning

Video face recognition is one of the most challenging topics in the field of image processing and computer vision and has a wide range of application value.Today although some promising results have been achieved in this area there are still serious problems in practical applications.Especially in the face recognition of video due to the lack of active cooperation of the detected person the detected faces often have large differences in observation angle light size position etc.which brings about the recognition work.Bigger challenge.For this reason this paper has carried out the research work of video face recognition technology based on deep learning.According to the various problems faced in video face recognition process it has designed and implemented an efficient face detection and recognition algorithm and restored super resolution.The technology introduced the face recognition process to further improve the image quality and improve the recognition effect.Finally a video face recognition system based on super-resolution restoration was built to verify the feasibility and effectiveness of the relevant algorithms and received good results.The main research content includes:A MTCNN-based video face detection algorithm is designed and implemented and compared with traditional face detection algorithms based on Harr features and AdaBoost.A video face detection algorithm based on MTCNN deep learning network was designed and implemented for the problems of pose illumination and occlusion in video face detection.The algorithm uses the inherent correlation between detection and calibration and cascades in depth.Improve the detection performance under the framework of the task;and further use the three-tier cascading architecture combined with a well-designed volume neural network algorithm to achieve a rough positioning of face detection and key points.At the same time based on the in-depth study and analysis of the traditional face detection technology a face detection algorithm based on Harr features and AdaBoost is implemented to perform the analysis with the aforementioned algorithm.Experimental results show that the MTCNN deep learning network method can better extract the effective features of the face in the video and achieve more accurate detection.Compared with the traditional AdaBoost and cascade structure-based methods the positive rate is Improved significantly.Propose a video face recognition algorithm based on multi-layer optimized LeNet model.In video face recognition how to improve the recognition rate for the captured face image is one of the research focuses.In this paper starting from the characteristics and actual needs of video face recognition a video face recognition algorithm based on a multi-layer optimized LeNet model is designed and implemented.Based on the traditional LeNet-5 network the video face features and recognition are specifically Demand a multi-layer optimization of the network structure an increase of a layer of convolution and pooling layers and redesign of the activation function parameter update rules back propagation methods and other parameters to optimize multiple data sets The test results show that the algorithm obtains a high recognition rate and the recognition rates in the two data sets of LFW and ORL reach 97.2% and 97.1% respectively.A face image super-resolution restoration algorithm based on a concise VGG(Visual Geometry Group)network is designed and implemented for low-quality video face recognition.Aiming at the poor imaging conditions and low resolution of face images faced by video face recognition a face image super-resolution restoration algorithm based on streamlined VGG and deconvolution is designed and implemented.Through the simplified design of the traditional VGG model the mapping relationship between high-and low-resolution images is modeled and then the deconvolution network is used to realize the super-resolution restoration process of the image.The performance of the algorithm is tested using the common data sets LFW and ORL.The results show that the algorithm's super-resolution recovery performance is significantly improved compared to the deep learning super resolution recovery network SRCNN(Super Resolution Using Conventional Neural Network)algorithm.Furthermore it is used in the recognition process of low-resolution face.The experimental results show that the recognition rate of the high-resolution face image restored by super-resolution is higher than that of the original low-resolution face image from 62.46% to 89.96%.The proposed super-resolution restoration algorithm improves the effectiveness of face recognition.Designed and developed a video face recognition system based on super-resolution restoration.Based on practical application requirements a prototype system for video face recognition based on deep learning was designed and developed.The feasibility and effectiveness of the proposed face detection recognition algorithm and super-resolution restoration algorithm were tested and verified.The Shilin DSP monitoring network camera is used as the video stream input and then the algorithm implemented in this paper is used to perform face detection face recognition and face image super resolution restoration processing on the loaded video stream and finally the recognition result is displayed.Finally through two sets of different angles of video surveillance tests the system's various functions have achieved the expected results and proved the effectiveness of the system.