Based On Level Set With Shape Constraint On Iris Segmentation Method Research

With the continuous development of information technology, traditional identificationmethods have been unable to meet the security requirements of modern society increasing dayby day. Biometric identification is safer identification technology in recent years. Among them,Iris recognition has become the focus in scientific research, industry and information securityfield for the advantage of its uniqueness, permanent, high reliability and non-invasive. Level setmethod on Iris segmentation is an extension of classical active contour model, which controlsinitial boundary evolution to the real object boundaries by partial differential equation. Thepaper introduces prior shape to constrain the curve deformation. The method can effectivelysolve edge blur and part boundary obscured or defected, and to further improve thesegmentation performance of active contour model.Aim at the research content, we study and improve a key technology of iris recognition,and establish a complete iris recognition system. The main contents are as follows:（1） Iris segmentation method is proposed based on improved level set. Using the irisprior knowledge, the method integrates circular shape energy into CV model by minimize thecircular parameters to segmentation internal boundary. However, external boundarysegmentation is introduced adaptive area term, which uses the property of the second derivative-passing zero to find boundary. And also using boundary gradient information makes sure thecurve evolution to stop accurately external boundary.（2） Feature extraction of2D-Gabor Filter. Analyze the character of2D-Gabor filter onlocation, scale and frequency. A set of Gabor filters with different scales are designed, whichcan well extract the overall and partial character of Iris textures. According to phraseinformation around sample points, unique iris code from different people is generated. （3） Feature extraction of Ellipse Gaussian Filter. With proper direction is designedbased on the scale changing law of iris texture in order to extract the region grayscale nearsampling points to generate code sequence. The hamming distance from different codesequence is calculated to make criteria for classification.（4） Experimental verification. Experiment results on iris database show that theproposed segmentation method about accuracy and speed is respectively99.39%and4.8s,which is more accuracy, quicker, robust than the traditional. Meanwhile, the proposed featureextraction in error rate is0.05%and matching speed is improved about10times. So Featureextraction and matching of elliptic Gaussian filter is higher recognition rate and fast computingspeed.