Research On Identity Authentication Algorithm Based On Lip Motion

Biometric technology has been widely used,according to the lip moving video(without sound)to carry out the identity of the person,compared with the current common identity authentication technology,which has the advantages of high uniqueness and low complexity.There are five important stages in the whole lip moving identification system: Face localization,lip localization,image preprocessing,feature extraction,identification and classification.In this paper,the algorithm of the above several stages is studied in detail,the image preprocessing,feature extraction,identification and classification algorithm are innovated and improved.In this paper,we use the algorithm of Adaboost and Haar feature to l ocate the face region,the feature vector extracted by the Haar feature template is put into the strong classifier of the Adaboost algorithm for face recognition.After determining the region of the face,in the face of the various organs the proportion has a relatively variable range of circumstances,the lips are coarsely positioned and ultimately get the ROI we need.For image preprocessing and feature extraction.First,this paper presents a video frame regularization algorithm,through the regularization of time,the inter-frame picture of the gray-scale changes in the largest picture extracted,thus the different length of the lip-motion video to the same length of regulation,so as to extract a representative lip moving pictures.The paper then proposes a method based on the SIFT keypoint,the texture and motion features are extracted by the changing pixels between the feature pictures,the motion vector and the moving direction of the key points.Finally,the feature vector is reduced by PCA algorithm,and finally a representative lip-moving feature is obtained for authentication.Finally,according to the characteristics obtained,a sorting algorithm is proposed by comparing the ratio between the eigenvalues and the threshold.And compared with the common image processing features LBP and HOG features under the same experimental conditions,the processing results are compared.The experimental results show the validity and feasibility of the proposed algorithm.