Preprocessing And Recognition Of Ocular Biometrics

With the continuous development of society,people pay more and more attention to the security of personal information,and biometric technologies are favored by more people as a stable and reliable identity authentication way.The eye area of human contains rich biometric information,which is mainly represented by iris,sclera and periocular,corresponding to three authentication methods of iris recognition,scleral recognition and periocular recognition,respectively.The preprocessing of biometric image is the most critical step for biometric recognition and plays a vital role in enhancing the effectiveness of biometrics.Based on the studies on preprocessing of human eye images,this paper analyzes the characterizations of different eye regions and strengthens the feature in a targeted manner to improve the accuracy and stability of human eye biometric recognitions.This article mainly studies the following works:(1)By analyzing the different feature areas of human eye and combining the preprocessing methods in different area feature recognition tasks,the data sets including key point detection data set of human eye,iris location and detection data set,iris,sclera and pupil full segmentation data set are established.All data sets are from different image acquisition equipment and experimental conditions.They have been manually selected and labeled,which can meet the needs of human eye biometric image preprocessing in most scenes.These data sets will provide a good data support for the realization of human eye preprocessing and eye feature analysis.(2)Eye location and state estimation is an important step in the task of eye feature recognition,which determines the selection of eye feature recognition methods.Human eye detection and state estimation are easily affected by the change of acquisition environment and light source,and the slight change of head posture will also lead to the decrease of the accuracy of human eye location and state estimation.Therefore,an eye location and state estimation method based on key points is proposed.By stacking multiple hourglass networks to predict the information of four key points of the human eye,a reasonable eye detection strategy and state estimation method are designed to locate the position of the human eye,and estimate the opening and closing state of the human eye,which has a good improvement in anti-interference ability and stability.(3)Using the periocular area to study the identification of the target.The periocular area gathers the most discriminative features of the face,which show the identity attribute information of the recognition object.In the biometric system,getting the gender attribute information of men and women is helpful to analyze the regularity of big data.Using CNN network to extract the characteristics of periocular region,and adding the eye membrane segmentation to increase the attention mechanism to enhance the contribution of discriminant features can improve the accuracy of periocular attribute recognition.