Preprocessing And Recognition Methods In Iris Recognition

Irises of human beings have characteristics of uniqueness, stability, irreversibility and anti-fraudulence, and so iris based biometric recognition technology has widespread applications in the fields of safety control and electronic commerce etc. With advance of iris recognition technology, iris recognition systems of higher precision are required. Two primary reasons that affect recognition rate are noises that occlude the iris regions and feature extraction algorithm.Eyelid and eyelash noises lead to erroneous iris encoding which severely affect accuracy of iris recognition. The thesis proposes a new method to detect the eyelid using non-parametric density of parabola. The proposed method has no need to detect edges, and considers both points in the model and in their neighboring regions. In addition, a method of breaking-point connection and region growing is presented to detect the eyelash, which considers the connection-, edge- and direction-characteristics, and meanwhile avoids bad effects of threshold selection. The traditional methods based on Hough Transform to locate coarsely human eyes are computationally expensive, and so a novel ellipse fitting method based on RANSAC is proposed that is efficient. To address the problem that traditional iris recognition algorithm is susceptible to the transformation of scale, rotation and affine, a new feature matching method based on SIFT is used to extract the textual features of iris which gets better result in very noisy database.