Low-quality Iris Image Recognition Algorithm

With the development of information technology, iris recognition has become a hot research topic in the cross-subject of Applied Mathematics, pattern recognition, image processing and information security. Recently, more and more researchers devote themselves to iris recognition, but the research of recognition on low-quality iris image (for example, covered by the eyelashes and eyelids, at-a-distance, motion-blurred, focusing-blurred, rotation, glasses, poor lighting etc.) is lacking. Most practical application systems require user’s strict cooperation to obtain high-quality iris image in the process of iris image collection, and most algorithms have certain requirements for the quality of iris image. Nevertheless, it is hard to ensure the quality of iris image in the practical application.To solve the problem of low-quality iris image recognition, through analyzing and researching low-quality iris image, an iris recognition algorithm which is based on multi-orientation2D Gabor and Adaboost is proposed in this paper. The main works are accomplished in the paper as following:1. The working principle and development of iris recognition technology were discussed, and several mainstream iris recognition algorithms were illustrated.2. According to the problem of existing algorithms, a new method of iris feature extraction and recognition was proposed based on the global and local texture of iris. First of all, according to whether pupillary segmentations were accurate or not, the irises were normalized based on pupillary and outer boundary or based on outer boundary. Then irises were divided into different amount of patches according to normalization. Furthermore multi-orientation2D Gabor filters were used to encode the whole iris and patches, and the feature vector was obtained. Finally, Adaboost algorithm was proposed to choose some features whose recognition performances were well for accurately and inaccurately segmented irises separately.In addition, Random forests and Fisher linear discriminant were used for recognition instead of Adaboost algorithm in this paper.The results of the experiments carried out on database CASIA-IrisV3-Lamp and noisy database Ubiris.v2with Matlab indicate that the proposed algorithm in this paper has a higher accuracy and robustness. What is more, the proposed algorithm has good performance on low-quality iris image, which overcomes the disadvantage of existing algorithms that can’t characterize low-quality effectively. In addition, the proposed algorithm was ranked2nd among all of the67participants from28different countries/districts in the international iris competition NICE.II.