Summary: | 碩士 === 國立暨南國際大學 === 通訊工程研究所 === 97 === In biometrics, iris recognition system has a high-level security. However, the pupil size
will change with different illumination and the iris texture deformation caused by pupillary
variations. So how to predict the deformation degree of the iris correctly is an important
issue. Yuan and Shi proposed a non-linear normalization model, with the prior definition
parameter $lambda_{ref}$ through the solution of two simultaneous equations to solve the iris deforma-
tion caused by pupillary variations problem. And it shows that their approach perfoms better
than linear normalization method. But this non-linear normalization model has a high com-
putational complexity. When the normalization of image size increases, the computing time
will increase rapidly. In iris recognition applications, how to achieve real-time conditions
is a problem, so it is must need fast calculation. This thesis proposes a fast algorithm us-
ing the law of cosines which can solve the non-linear normalization model simply and fast.
The exprimental results show the computing time reduce 100 ms, and the equal error rate
(EER) decrease 0.94% which compare with linear normalization. This thesis improves the
non-linear normalization of speed, and shows performance better than linear normalization.
|