Rotation Compensated Human Iris Matching

By introducing rotation compensation into a human iris matching system, improved matching is demonstrated while reducing both the computational complexity and storage requirements. This is achieved by using Fourier domain cross correlation to estimate the relative rotation of two iris images. This eliminates the need to store and compare codes from multiple orientations of the same image. Instead, only one code is stored, plus the Discrete Fourier Transform of an annular segment derived from the iris. In addition, the cross correlation function can be used as a biometric either on its own or in combination with other iris codes. A Peak to Sidelobe Ratio metric is used to discriminate matching and non-matching correlation functions. Used to preselect irises for matching, the method greatly reduces feature-vector comparisons. Alternatively a cascaded classifier approach can be used, with the correlation stage tuned to make no false rejections, followed by a Hamming distance metric on a standard iris code. The performance and robustness of the new technique are tested on the CASIA iris image database synthetically altered to generate rotated, noisy and randomly occluded normalized iris images. Improved matching performance is achieved with a 21 times reduction in matching