Iris Recognition: An Analysis of the Aliasing Problem in the Iris Normalization Stage

Iris recognition has been increasingly used with very satisfactory results. Presently, the challenge consists in unconstraining the image capturing conditions and enable its application to domains where the subjects' cooperation is not expectable (e.g., criminal/terrorist seek, missing children). In this type of use, due to variations in the image capturing distance and in the lighting conditions that determine the size of the subjects' pupil, the area correspondent to the iris in the captured images will be highly varying too. In order to compensate this variation, common iris recognition proposals translate the segmented iris image to a double dimensionless pseudo-polar coordinate system, in a process known as the normalization stage, which can be regarded as a sampling of the original data with the inherent possibility of aliasing. In this paper we analyze the relationship between the size of the captured iris image and the overall recognition's accuracy. Further, we identify the threshold for the sampling rate of the iris normalization process above which the error rates significantly increase