Robust pre-processing techniques for non-ideal iris images
Purva M. Barve · 2005
The human iris has been demonstrated to be a very accurate, non-invasive and easy-to-use biometric for personal identification. Most of the current state-of-the-art iris recognition systems require the iris acquisition to be ideal. A lot of constraints are hence put on the user and the acquisition process.;Our aim in this research is to relax these conditions and to develop a pre-processing algorithm, which can be used in conjunction with any matching algorithm to handle the so-called non-ideal iris images. In this thesis we present a few robust techniques to process the non-ideal iris images so as to give a segmented iris image to the matching algorithm. The motivation behind this work is to reduce the false reject rates of the current recognition systems and to reduce the intra-class variability. A new technique for estimating and compensating the angle in non-frontal iris images is presented. We have also developed a novel segmentation algorithm, which uses an ellipse-fitting approach for localizing the pupil. A fast and simple limbus boundary segmentation algorithm is also presented.