Research On Lip Movement Identity Authentication Technology For Specific Person

Because of its uniqueness and convenience,biometrics is favored by people in daily life.Lip movement identity authentication technology is a new recognition technology that integrates individual physiological characteristics and pronunciation habits.This combination makes the uniqueness of the technology better and the user requirements are relatively low and hardly falsified.The goal of this dissertation is to identify the specific speaker based on the change of lip during speech.Lip database is the basis of the whole study.In this dissertation,a comprehensive analysis of the existing static face database and bimodal corpus database at home and abroad is shown.800 face images in the CAS-PEAL database and 800 non-lip images in the ORL database are composed into the static lip database.At the same time,a self-built small-scale corpus database is constructed with reference to the construction of existing databases.Taking into account the shooting angle of the database is aligned with the face,the designed lip positioning process is face positioning,lip rough positioning,lip refinement.This dissertation adopts the Haar-based Adaboost algorithm to realize face detection,which has a good recognition effect and real-time performance.The rough positioning of the lip is based on the distribution of facial features.The fine positioning design process is thresholding,opening operation,filling connected domain and smoothing.The experimental results show that the location shows well whether the lip is closed or open.Aiming at the problem of large amount of data when the video is converted into static pictures,this dissertation proposes an improved method of selecting the representative picture sequence.The beginning and the end are determined according to the lip contour feature,and the representative images are selected according to the inter-frame difference method based on the maximum gray level variation.Experiments show that this method is effective and can achieve the purpose of refining the database.Considering that the accuracy of face positioning and lip positioning often can't achieve 100%,this dissertation designs a pattern recognition method based on lip static feature and neural network,and compares the Hu feature,GLCM feature and GGCM feature.In this dissertation,a personal lip recognition based on PCA is proposed.Establish the characteristic lip space and identify the identity by measuring the distance between the vector projected to the lip space and the training sample by.In addition,in order to enhance the flexibility of the system,it is proposed to store the lip feature in the form of a two-dimensional code so as to share the password and authorize it.Then,this dissertation designs a specific human lip recognition method based on the SURF algorithm.According to the improved feature point matching method,the speech content is identified according to the total number of matching points between two sequences of images.Experiments show that this method performs well.