Wavelet Network Based Approach for Iris Localization

This paper proposes a novel algorithm for detecting inner and outer boundaries of iris images using a histogram thresholding segmentation scheme for detecting the first one and a radial browsing segmentation for the second one. The former segmentation method is ensured think to a usual technique as known: the inter-means algorithm. The iris inner boundary delimits the pupil and allows us to find out its centre that serves us soon after. The later segmentation method consists in detecting only eight points to estimate the outer boundary of the iris. Indeed the iris can be delimited by four circle bows: one from each side. The right and left bows are both centred on the pupil centre but with different rays. So, one point for each bow is sufficient to perfectly determine it. The upper and bottom bows have their centres elsewhere, then, they need three points each one to be perfectly definite. To find out these points we study the variation of the grey level intensity according to eight directions leaving from the pupil centre. The point belonging to the outer boundary has the particularity to be lightly darker than their neighbourhoods. This particularity is mathematically traduced by being a sink. Such a point can be deduced only from a continuous and twice differentiable function in the whole of its points. A wavelet network estimator is then used to approximate grey level intensity vector in order to generate the required function. This localization method has been tested and successfully evaluated on 756 iris cases