Research And Implementation Of Face Recognition Based On Semantic Features

At present,the face recognition technology is developing rapidly and becoming more and more mature.And it is more and more widely used in people's daily life.However,in the complex environment,the face recognition system still has the problems of face false detection,low accuracy and low robustness,which cannot meet the actual business requirements well.Technically,it is mainly due to the difficulty in extracting discriminative features and paying too much attention to global features while ignoring the importance of local semantic features.To solve these problems,this thesis innovatively introduces face semantic segmentation technology,proposes a method for face false detection filtering based on semantic features and a face recognition algorithm based on the fusion of global and local semantic features,which greatly reduces the face false detection rate and enhances the accuracy and robustness of face recognition.The main contributions are as follows:(1)Aiming at the problem that the existing schemes cannot extract representative and discriminative features well,a method for face false detection filtering based on semantic features is proposed.By introducing face semantic segmentation technology and using feature engineering technology,a face semantic feature with stronger representation ability is constructed.Then combined with the advantages of simple structure and strong generalization ability of support vector machine,the effective classification and screening of false face detection results is realized.It further reduces the face false detection rate and improves the performance and robustness of the face detection algorithm.The experimental results show that this method can effectively reduce the face false detection by about 80% and improve the average accuracy of the overall face detection algorithm by about 1%.(2)Around the problem that existing algorithms mostly focus too much on global features and ignore the importance of local semantic features,a face recognition algorithm based on the fusion of global and local semantic features is proposed.This thesis uses face semantic segmentation technology to locate the local semantic area of the face,and then perform feature extraction on the global and local semantic areas respectively.Therefore,the local semantic features of the face are explicitly introduced.After that,we perform effective feature fusion on global and local features with local occlusion information.It improves the representation ability and robustness of the final features,which further improves the accuracy and robustness of the face recognition algorithm in complex scenes.The experimental results show that the method performs well in a variety of face recognition scenarios,especially in scenes with partial occlusion,which has an accuracy improvement of more than 1.67% compared to other general algorithms.So this algorithm has better face recognition results and robustness,and is more in line with business application requirements in complex environments.(3)A complete face recognition system is designed and implemented.The method for face false detection filtering and face recognition algorithm proposed in this thesis were applied to specific face recognition business scenarios.The system shows good performance in both face comparison and face search.