Video Face Recognition Based On Deep Neural Network

With the rise of data mining and artificial intelligence in recent years,plenty of major breakthroughs have been made in the field of biometrics.As an important field of biometric technology,face recognition is already playing an important role in our lives.However,even the field of face recognition has been widely studied by a large number of researchers,video-based face recognition is still a challenging field of study which requires more in-depth exploration.Today,deep learning has been widely used in the field of face recognition.In this paper,a video face recognition model based on deep neural network is proposed.This approach possess high recognition accuracy and a strong real-time performance.This article mainly includes the following research contents:(1)The multi-task cascade convolutional neural network is used to implement the face detection function.This paper uses multi-task cascade convolutional neural network as the deep neural network model of the face detection part.The network task of the entire neural network is a combination of three smaller tasks: face and non-face classification,face bounding box regression and facial landmark localization.(2)Face feature extraction function is implemented by using a model based on deep convolutional neural networks.A deep learning based algorithm for face recognition is designed for facial feature extraction from video images in this paper.Some of the key parts,such as network structure and loss function are also optimized for the video image in natural scenes because of their more complex features,which makes the network more suitable for the needs of video face recognition in natural scenes.(3)Center loss is used to shorten the distance between features of the same class.In this paper,the face recognition model uses the central loss approach instead of the triple loss to achieve the purpose of aggregating similar features,so that the distance between features of the same category could be minimized.(4)The face re-location algorithm is proposed to improve the real-time performance of face recognition.In order to ensure that the face image in the video can be processed quickly enough,a face relocation method is proposed in this paper to reduce the computational cost of the face recognition process.Finally,the performance of face recognition algorithm is analyzed,and compared with other face recognition algorithms based on deep neural network.The experimental results show that the accuracy of classification of the video-based face recognition algorithm reaches 96.0% on the You Tube Faces dataset.In addition,when it comes to videos with the resolution of 1024x576,the processing speed of the face recognition method with the usage of the Nvidia GTX 970 can reach approximately 20 frames per second.Finally,the face recognition algorithm is tested with the multi-face classification task in natural scene videos and images.The experimental results show that the model is robust to face recognition under complex conditions.