No-reference Iris Image Quality Assessment

With an increasing emphasis on information security, as a high performance identity recognition algorithm, iris recognition has been received extensive attention. This paper focuses on the iris image quality assessment in iris recognition and the main works are as follows.(1) The pupil rough location and fast iris image quality assessment algorithm based on radial symmetry transform was firstly proposed. The accuracy of the pupil location algorithm is 100% in Chinese Academy of Sciences CASIA-IrisV4-Thousand database. In addition, because the radial symmetry transform used the circle’s integrity information and the gradient of pixels in circle’s edge, a threshold was set for the votes of radial symmetry transform to fast judge non-iris image and detect eyelash or eyelid occlusion and seriously blurred iris images.(2) For the general blurred iris image, an iris image clarity evaluation algorithm was proposed, which was based on the spatial domain statistics of iris regions. The sharp edges are the favorable information to judge an iris image whether blurred or not. But in iris images, there are too noisy in the sharp change regions. This paper firstly used the result of rough pupil location to select the iris region with pupil edges. This operation eliminated the influence of skin and eyelashes as far as possible. Secondly, the pupil zooms because of the light intensity change, so the iris regions were normalized. Lastly, the empirical distribution curves of the iris’s local Gaussian weighted normalized information exist a Gaussian-like appearance and they are significant differences in clarity and blurred iris regions. A generalized Gaussian distribution was used to exact an 18-D feature vectors and support vector machine (SVM) was used to train a clarity and blurred iris image classifier. The result proves that this classify algorithm’s accuracy is 96.8%. It proves that the no-reference iris quality assessment, that based on radial symmetry transform and iris spatial domain statistics, is robust and calculate fast. The algorithm can effectively wipe off the low quality iris images which affect the recognition and has an inspiring practical application.