Classification-based algorithms for iris outer edge location

Algorithm design of iris outer edge location is one of keys in iris recognition. It dose ask for an excellent performance as well as an acceptable running time. Low-level technologies and simple classifiers are employed in this paper to meet these needs. First of all, we propose a gradient computation for point detection to ensure a shorter running time. In this part, points can be located as many as possible. Then classifiers distinguish points on fake boundaries and iris boundaries. The number of points will be dramatically cut down. One classifier bases on the characteristic shape of irises, one bases on the distances between points and pupil´s center and the other bases on gray gradient of each point. At last, LSE (least square method) is employed to fit a curve of iris boundaries. The algorithm has been verified by experiments. The correct rate exceeds to 99% for public databases CASIA, while the running time is shortened to about 80 ms.