S.-W. Lee and S.Z. Li (Eds.): ICB 2007, LNCS 4642, pp. 790–799, 2007. © Springer-Verlag Berlin Heidelberg 2007 Shape Analysis of Stroma for Iris Recognition ∗ S. Mahdi Hosseini 1 , Babak N. Araabi 1 , and Hamid Soltanian-Zadeh 1,2 1 Control and Intelligent Processing Center of Excellence, School of ECE, Univesity of Tehran, P.O. Box 14395-515, Tehran, Iran 2 Image Analysis Lab., Radiology Dept., Henry Ford Health System, Detroit, MI 48202, USA sm.hosseini@ece.ut.ac.ir, {araabi,hszadeh}@ut.ac.ir Abstract. In this paper, a new shape analysis approach for iris recognition is proposed. First, the extracted iris images from eye portrait are enhanced by image deblurring filter which computes restoration using FFT-based Tikhonov filter with the identity matrix as the regularization operator. This procedure produces a smooth image in which shape of pigmented fibro vascular tissue known as Stroma is depicted easily. Then, an adaptive filter is defined to extract these shapes. In the next step, shape analysis techniques are applied in order to extract robust features from contour of the shapes such as support functions and radius vectors. These features are invariant under iris localization and mapping. Finally, a feature strip code is defined for every iris image. Introduced algorithm is applied to UBIRIS databank. Experimental results show efficiency of the proposed method by achieving an accuracy of 95.08% on first session of UBIRIS. Keywords: Biometric Recognition, Stroma, Tikhonov Filter, Shape Analysis. 1 Introduction Literature of iris recognition is dominated by wavelet methods. First method was proposed by Daugman [1, 2] which used multiscale quadrature wavelets to extract texture phase structure information of the iris. Ma et al. [3–4] adopted a well-known texture analysis method (multichannel Gabor filtering) to capture both global and local details in iris. Wildes et al. [5] with a Laplacian pyramid constructed in four different resolution levels and the normalized correlation for matching designed their system. Tisse et al. [6] combined the original image and Hilbert transform to demodulate the iris texture. 2D Haar wavelet was used by Lim et al. [7] and applied an LVQ neural network for classification. Kumar et al. [8] developed correlation filters to measure the consistency of iris images from the same eye. The correlation filter of each class was designed using the two-dimensional Fourier transforms of training images. Bae et al. [9] projected the iris signals onto a bank of basis vectors derived by independent component analysis and quantized the resulting projection coefficients as features. Gu et al. [10] used a multiorientation features via both spatial ∗ This paper is for BBSPA competition.