Iris Template Protection Based On Face Feature Information

With the development of biometric technology,the biometric template protection becomes a new advanced frontier.Iris is a stable and widely used biometric,and its original feature template is very important because of the uniqueness of iris feature.Once the original iris template is stolen,it will lead to the leakage of users' privacy and seriously affect the security of the users' personal property.Therefore,the protection of the original iris template has become one of the main research hotspots in this field.In this thesis,the main research content is the method of the iris template protection.Aiming at the problem of iris template protection,three iris template protection methods based on face feature information are proposed: two iris template protection methods based on eye distance ratio and an iris template protection method based on face feature.In the pre-processing stage,the classical algorithm is used to locate and normalize the iris,and then the Log-Gabor transform is used to extract the iris feature.In the stage of template protection,the first method and the second method use the user-defined eye distance ratio as the encryption key,and the third method uses the face feature information as the encryption key.In the first method,it uses hierarchical Logistic chaos mapping and Bloom filter to protect the iris template based on the eye distance ratio,and uses Hamming distance classifier to classify;in the second method,it uses random projection and Logistic chaos mapping to protect the iris template based on the eye distance ratio,and uses Euclidean distance classifier to classify;double Bloom filter and discrete logarithm are used to protect the iris template based on face feature in the third method,and uses Euclidean distance classifier to classify.Due to the iris is not related to the face,this thesis uses two combinations: CASIA iris image database(version 1.0)+ CASIA iris distance and CASIA iris V3 interval + CASIA iris distance to simulate and evaluate the performance of the above methods.Randomly select 50%samples of each person in iris databases as training set,and the other as test set.The recognition accuracy of the three algorithms in the two combination ways are 98.15%/96.56%,97.22%/95.21% and 97.22%/95.23% respectively.Compared with the unencrypted method,the recognition performance is poorer,but the security performance is improved.In addition,the algorithms in this thesis meet the three standards of iris template protection method,which are irreversibility,revocability and unlinkability.To sum up,the methods proposed in this thesis not only meets the high recognition performance,but also realizes the protection of the original iris feature template.