Based On Multitask Cascade Convolutional Neural Network Face Recognition System

Face recognition technology has developed rapidly on the basis of the rapid development of computer vision communication technology.With its high recognition efficiency,the traditional recognition algorithm is far behind,making it slowly become an image.In recent years,with the advent of big data technology and the great enhancement of computer hardware conditions,image acquisition and processing capabilities have been significantly improved.Face recognition technology not only has accurate recognition effect on faces in complex environments,but also specific gestures and Multi-angle joint faces can also be quickly identified.Moreover,compared with the biometrics currently used for identification,its advantages such as convenience,high intelligence,human-machine friendly,security and stability make the face recognition technology become a popular technology in the field of identity recognition.From the aspect of multitasking,the system design of face recognition is based on two core tasks to guide the design of the system framework including the face detection task module and the face recognition task module.The face detection task module further includes functions such as face detection,key point positioning,and face alignment.The face recognition task module mainly functions such as face feature extraction and face feature comparison.The various modules not only complement each other,but also stand out and have a strong independence.The technique for detecting human faces is based on the independent architecture of Anchor-based cascading structure.The face recognition task module uses the Inception-Resnet feature extraction network architecture based on Tensorflow framework.Finally,the cosine distance of the facial features is calculated to perform face matching.The face recognition model is pre-trained by the means of offline training.The face data about the laboratory is collected,and then the migrated model is learned.All systems are run one by one,and the overall performance test is carried out.The test results prove that the various task modules designed in this paper have achieved good results.