The Noise Detection For Iris Recognition

With the development of the informational, digital and networked society, the security to the world and social life is put forward increasing requirement. In this condition, the traditional security technique presents the unresolved and severe faults. Moreover the biometric recognition technique is the best method to improve the security of the informational, digital and network society. The iris recognition has a satisfying performance due to its uniqueness, stability, high reliability and non-invasion. Statistically, the error rate of iris recognition is the lowest among all the biometric recognition technique.In an iris recognition system, the major research object is the iris texture. However, under the high reliable system, the captured iris image contains not only the iris information, but also the sclera, eyelid, eyelash, reflection, and so on. If the disturbed factors are recognized as the iris feature into the iris image coding, there will be false iris code. Therefore, this can greatly affect the performance of iris recognition. This paper analyzes different noises in the iris region and takes different methods to eliminate different noises, so we can avoid the noise region in the feature extraction and matching, meanwhile improve the accuracy rate of iris recognition.By observing the extensive experiments, the noises of the iris region are mainly composed of eyelid, eyelash and reflection. Then different methods are proposed to detect these noises in this paper. The eyelid noise contains the upper eyelid and the lower eyelids, the edges of which are close to parabolic, the traditional method that combines horizontal edge detection with Sobel operator and parabolic Hough transform are used to find the upper and lower eyelids;Eyelash is composed of the upper and lower eyelash. The eyelashes near the lower eyelid are sparse and smallish, whose projection in iris region is little. Moreover the projection of the eyelash near the upper eyelid on iris region is much. According to the eyelash characteristics, first the local intensity minimum algorithm is used to obtain the candidate points. Then use the information of the eyelash length and location position to remove the false eyelash points. Finally a slice thatcontains the eyelashes is denoted as eyelash noise region. The reflection detection can be classified into two steps: (1) obtain the points of higher intensity, which belong to reflection and eyelid;(2) they are identified as points in a reflection or in an eyelid, according to the characteristic of lower intensity around the reflection. Extensive experimental results show that the algorithm is effective, which not only has high precision, but also has good robustness.