External Feature Extraction And Application Research Of The Eyebrow

At present,it is generally considered that the authentication method of higher security is to use of the unique characteristics(such as fingerprint,face,etc)of biological itself for identification.This way has been widely used in financial service?video surveillance?information security,human-computer iteration,criminal identification,electronic commerce?as well as in areas of the entry and exit management.In view of human eyebrows meets the use of biometric identification with features of universality,uniqueness,stability and the ease of collectability.Eyebrows can be used as a means of identification.Eyebrows have distinct contour and texture features,but the current research of eyebrows mainly stays on the single eyebrow image,and lacks the related research on the external features of eyebrows,therefore,the further study of the eyebrow has great significance in the field of biometrics.The main contents of the research work in this paper are as follows:(1)To extract the eyebrow contour,this paper briefly describes the basic principle of the level set algorithm firstly,and then the level set model based on bias field correction(Li's model)is focused,which is suitable for the segmentation of images in the presence of intensity inhomogeneity,but often the initial curve is set unreasonable,which leads to the evolution time is too long,especially,the part of initialization is improved in this paper,and a pseudo-sphere-based edge detector is introduced,coupled with morphology of the closed and filling operations,to realize the coarse position of the initial curve,the initial curve of the evolution is automatically set to the initial contour near to the region of interest(eyebrows' edge),the level set evolution time of Li model is greatly reduced.(2)A pseudo-sphere-based edge detector and Li's model are integrated,the eyebrow contour of the pure eyebrow image is obtained by level sets evolution,on the basis of it,the geometrical features of the eyebrow is used to calculate shape features and directional features,then the gray level co-occurrence matrix(GLCM)is used to calculate texture features of the eyebrow,the external features model of eyebrows is built based on feature vector.(3)Experimental results show that the eyebrow contour based on our model is more accurate than Li's model under the same iterations;in a self-built nature eyebrow image library(100 persons),the matching rate of identity verification is as high as 90.59%,the eyebrow recognition rate of external features model for the single one can reach 86.1%(similar with HMM and 2DPCA),and 90.2% for the double eyebrows.Even in the eyebrow library without difference in heavy-light,based on the model of shape and direction features,the recognition rate of single and double eyebrows is 88.1 and 88.7%,which indicates that eyebrows can alse be recognized.