Research And Implementation Of Face Recognition Enhancement Algorithm Based On Multi Task Cooperation

In recent years,deep convolution network has achieved high accuracy in face recognition and attribute classification tasks.Although these two tasks are closely related,it is still difficult to use their correlation to improve their accuracy.This thesis focuses on using multitask learning to do these tasks,and designs a multitask learning model.The model first uses the global information and identity information from face features to improve the attribute classification results,and then uses the semantic information contained in attribute features and a center based metric learning algorithm to enhance the ability of face recognition.The main work of this thesis is as follows:1.A face attribute classification one-way cross stitch network based on face recognition model is designed.For the problem that the attribute classification model lacks sufficient feature interaction with the face recognition model,this thesis extracts face features from different receptive fields for different scale face attributes,and designs a one-way cross stitch structure to integrate face features and attribute features.Compared with many algorithms,this model can output high-precision attribute classification results on the premise of ensuring the accuracy of face recognition.2.A face recognition enhancement algorithm based on multi task dynamic routing and center metric learning is proposed.Focusing on the problem that the importance of different face attributes varies greatly for different faces,this thesis designs an attention structure based on dynamic routing algorithm,which can calculate the weight of attributes to face from the adaptive algorithm.At the same time,by analyzing the problems of face recognition algorithm in the real scene,a center based metric learning algorithm is proposed,which further enhances the face representation ability of the model.The experimental results show that the algorithm is effective.3.Design and implement a face recognition and attribute classification system.By embedding the related algorithm proposed in this thesis,the system can simultaneously perform the task of face attribute classification and face recognition with high accuracy.This thesis introduces the total architecture of the system and the specific implementation of each module,and shows the actual operation effect of the system.