Research Of Biometrics Application Based On Scale Space Theory

Biometrics has been widely used in many areas, in which iris recognition has been well studied for its high uniqueness, low error rate, stable performance and so on; also, finger vein recognition is a burgeoning technology in biometrics, because it is free to interference, easy to capture and has low cost in device. Currently, the most methods are usually based on cooperative databases, and they may have degradation in performance in non-cooperative databases. For example, in iris recognition the scale of images may change with different capture distances and in finger vein recognition the images rotation may caused by different placement of fingers. So, how to solve these problems becomes a subject need thoroughly to probe.In this paper, we overcome problems in image rotation, translation and variant scales in finger vein recognition and iris recognition based on the SIFT algorithm:(1) Firstly, we summarize the recent development in iris recognition and finger vein recognition, and then we analyze the problems in related works. Then we introduce the common frameworks of finger vein recognition and iris recognition and analyze the basic algorithms in the frameworks.(2) Secondly, we introduce the scale space theory and the classical algorithm-SIFT, and by analyze the advantages of it we can apply it in the following works.(3) We analyze shortages of the related work in iris recognition based on SIFT, such as feature loss and noise introduction, which leads to a low recognition rate of system. Then we proposed a new iris recognition based on SIFT to solve these problems. Experimental results show that the proposed method can reach a higher performance compared with other SIFT based methods in CASIA database.(4) We first introduce SIFT in finger vein recognition to solve the performance reduction caused by rotation. First we select suitable parameter of SIFT based our finger vein database. Then we compare our method with binary pattern based method in non-rotated database and rotated database, respectively to verify the efficiency of proposed method. Experiments show the proposed method can reach a satisfying recognition rate in finger vein, which has a lower EER at1.71%compared with LBP at2.67%. Also, the proposed method has a stable EER at2.6%in rotated database while the performances of LBP degrade badly.