Research On Detection And Localization Of Pan-iris

With the advent of the network information age,the use of biometric identification technology to achieve the security authentication of personal identity information has become the focus of social attention,among which,the uniqueness,stability,non-invasion and security of iris recognition make it a leader in this field.Iris recognition is a technology to complete identity recognition by extracting iris texture information,which usually includes image acquisition,iris image preprocessing,iris feature extraction and coding,iris feature matching recognition.Among them,the acquisition of iris effective region,that is,the accurate iris texture,has become the decisive factor of iris recognition performance.Therefore,the related research related to iris effective region acquisition is defined as pan-iris detection,which mainly includes iris detection,eyelid detection,eyelash detection and so on.So far,the key and difficult problem of pan-iris detection is how to use detection methods to solve the interference factors of iris texture information.The eyelids and eyelashes are the main interference sources of iris texture,so the eyelid detection and eyelash detection selected in this paper are studied deeply.The traditional methods of eyelid detection can be divided into two categories,one is straight line fitting to detect eyelids,the other is parabolic fitting to detect eyelids.The detection of eyelid by straight line fitting has the advantages of low computational complexity and fast detection speed,but the detection accuracy is low,and even lead to the loss of a lot of iris information.However,in the process of detection,eyelash occlusion has become an important factor affecting the accuracy of eyelid detection due to the physiological structure of human eyes.Therefore,how to eliminate eyelash noise without weakening the edge information of the eyelid has become the most important problem to accurately detect the eyelid.According to the orthogonality of eyelashes and eyelids in the extended direction,a fast eyelid detection algorithm based on strong direction weighted Gaussian edge detection and RNL goodness-of-fit is proposed in this paper.Firstly,the upper and lower eyelids are segmented with the horizontal line of the center of the iris,and the region of the eyelids is roughly determined,and according to the characteristics that the eyelashes are approximately perpendicular to the eyelid structure,a smoothing filter with direction selectivity is designed,and the edge information of the eyelid is preserved while filtering the noise such as eyelashes.Next,a group of horizontal edge detection operators are designed,and according to the gray contrast degree of the upper and lower eyelid regions in the iris image,the edge detection operator is dynamically selected to extract the eyelid edge information in the iris image;Then,the candidate points of the eyelid edge are iteratively screened by the goodness of fit RNL;Finally,the least square parabola fitting is used to complete the eyelid detection.The existing research on eyelash detection basically adopts threshold segmentation or morphological methods.Although these methods can achieve a certain degree of detection effect,it is difficult to strike a balance among false eyelash detection rate,false non-eyelash detection rate and detection speed.Considering the characteristics of the eyelash,this paper puts forward the idea of dynamic threshold and layered detection,which uses different detection methods to complete the detection in a hierarchical and progressive way from the root of the eyelash to the tail.First of all,the eyelashes to be detected are mainly distributed in the upper half of the iris and almost none in the lower half,and there is a significant difference in texture gray distribution in the upper and lower half of the iris.Therefore,the center of iris circle is used to divide the upper and lower regions of the iris into equal arc blocks,and the upper and lower regions are compared with the discrete degree of gray distribution,so as to obtain the maximum distribution range of eyelashes in the upper half of the iris.Secondly,according to the large difference between the gray value of the root of the eyelash and the iris texture region,the minimum intra-class coefficient of variation method is proposed to realize the threshold segmentation.Finally,because of the diversity of the tail direction of each eyelash,and the difference between the gray value of the tail eyelash and the iris is relatively small,therefore,this paper combines the weighted multi-scale composite window proposed by Li Haiyan and the fingerprint block direction estimation method based on weighted linear projection analysis proposed by Bian Weixin to determine the direction of tail eyelashes.According to the design principle of directional filter and the shortcomings of existing literature in filter bank design,a directional filter with flexible multi-direction conversion is constructed to realize image enhancement,so as to complete eyelash tail detection in different directions.The samples from CASIA-V1,CASIA-V3 and BEE iris database are used for comparative experiments.the results show that compared with the eyelid detection by typical Hough transform and parabola fitting based on morphology and least square method,the fast eyelid detection algorithm with strong direction weighted Gaussian edge detection and RNL goodness of fit improves 1.7%and 0.4%respectively.The average detection time has been reduced by 99.4%and 97.5%respectively,and the average detection speed has reached 0.04±0.02s.Compared with the eyelash detection algorithm based on Gabor filtering and regional gray variance and based on eyelid contour and local gray minimum,the accuracy of eyelash detection algorithm based on variation coefficient and gradient weighted directional filtering is improved by 5%and 2.6%respectively,the average detection time is reduced by 61.1%and 50%respectively,and the average detection speed is up to 0.3±0.2s.