Research On Face Recognition Algorithm Based On Single Front View

Face recognition is a very important prat of biometric technology,which has a wide range of applications in criminal identification,background monitoring,access control security system and so on.However,in the video surveillance scene,when identify the face of pedestrians.Because of the pedestrian walking lead to status keep changing,and the crowd inevitably tend to obscure each other.What is more,if only a front view of the human face is stored,you need to rely on this single image to identify the target face in the video,the difficulty will be greatly increased.Due to the richness of the same person's face postures and expression,the use of machine learning method is effective,but the machine learning method needs to provide a large number of samples,accordingly,it will be a very big challenge for the task requirements that can only provide a single face image.In view of this problem,this paper has carried on the thorough research to the face recognition algorithm based on single front view.In order to solve the problem of insufficient training samples,this paper uses the method of homography transformation to generate multi-pose samples with the given single front view of face,and increase the number of training samples.The face recognition problem studied in this paper divided into face recognition of static scene and face recognition in dynamic scenes.In the face recognition of static scene,Gabor wavelet transform is used to extract the feature of face,and the Gabor wavelet coefficient are down-sampled by principal component analysis(PCA),the feature is easy to classify and improve the recognition speed.The classifier uses a support vector machine with a good sorting ability for small sample problems.The recognition of face in dynamic scene is realized by combining the method of static face recognition with moving target estimation and moving target tracking,The calibration of the test has a very significant effect on the leakage recognition and face error caused by partial occlusion or deflection angle.The experimental results verify the effectiveness of the proposed method.