Accurate iris segmentation based on novel reflection and eyelash detection model 论文

摘要

The authors propose a novel noise detection model for accurate segmentation of an iris. Eyelash, eyelid and reflection are three main noises. Eyelid has already been solved by a traditional eye model; however, eyelash and reflection have not been tackled. To determinate a pixel in an eyelash, the model presented follows the three criteria: 1) separable eyelash condition; 2) non-informative condition; and 3) connective criterion. The first and second conditions handle separable and multiple eyelashes respectively. The last criterion avoids misclassification of strong iris texture as a single and separable eyelash. For reflection, strong reflection points are detected by a threshold and the weak reflection points around the strong points are determined by connective criterion and statistical test. A number of images are selected to evaluate the accuracy and necessity of our noise detection model and the results are encouraging.