DETECTION OF ABNORMAL IRIS AUTHENTICATION

Abstract

This paper deals with detecting abnormal situations in iris authentication, such as cosmetic contact lenses, printed iris images, and fake eyeballs. The proposed algorithms exploit liveness of iris images and texture features to discriminate the fake iris inputs. In the proposed process, the liveness of iris images is first evaluated by inspecting the variation of eyelid curvature, blinking, and pupil gaze. The liveness test filters the most of fake eyeball and printed iris images. Then, the features of true iris images are examined to detect cosmetic contact lenses and fake iris images which have passed the liveness test. According to the experiments with commercial mobile devices, the proposed algorithms detect abnormal iris inputs with 95% accuracy. The proposed system prevents fake and inappropriate iris inputs, thus it makes iris authentication systems to be more reliable.

Existing System

Weighted LBP (Linear Binary Patterns)

Proposed System

This paper proposes some algorithms to detect abnormal iris inputs based on the characteristics of abnormal iris images described before. The detection is processed in two steps, one is the liveness detection, and the other is abnormal iris image detection using image features. First, the iris region is extracted from the input image using the Daugman’s algorithm, and the eyelid curvature is calculated using the edge detection method. And liveness detection is performed to detect abnormal iris that does not move, or blink eyes as shown in the printed iris images or fake eyeball. Liveness detection uses the variation of curvature of eyelid to determine whether the input iris image is real and alive. Unlike a normal iris whose eyelid shape varies from time to time, the curvature values of printed iris image or fake eyeball are almost constant. Therefore, abnormal iris can be filtered out first in this process. Next, the abnormal iris detection is executed to detect cosmetic contact lenses, printed iris images and fake eyeball that were not filtered. This method uses the average pixel intensity value in each part after dividing the extracted iris region into 16 equal parts in the radial direction. In detail, the probability models corresponding to the normal iris and the abnormal iris are calculated using the minimum value and the brightness change rate. This rate is the ratio of maximum and minimum values from the average pixel intensity values for the calculated 16 parts. Then, by combining the two probability models to maximize the Kullback-Leibler divergence between true and fake iris images, a discrimination model is generated, and the abnormal iris is determined for the input image.

Architecture

BLOCK DIAGRAM