Research Of Iris Segmentation Algorithm Based On Explicit Shape Regression

Iris recognition has great advantages on security and stability,and it has been used in many fields of life.Iris segmentation is a very important step in the iris recognition system,but existing algorithms have some problems in complex scenes and cross-device applications.By studying the process of iris segmentation,an iris segmentation algorithm based on explicit shape regression is proposed.The main work is as follows:In order to improve the robustness and generalization of the iris segmentation algorithm,an initial segmentation algorithm based on the iris shape feature points is adopted.Firstly,the iris shape feature points are located by the explicit shape regression method.The.iris shape is described by pixel-difference features,and the position of the feature points is updated step by step with the two-level cascaded regression strategy.Then,the accuracy of iris shape feature points localization is ensured by Gaussian distribution model combined with multiple sets of shape regression results.Finally,weighted least squares fitting method is used to get the initial segmentation results of iris image by fitting the upper eyelid boundary,lower eyelid boundary and the outer boundary of the iris.In view of the deviation of the iris outer boundary fitting result and the pupil coarse location,the coarse-to-fine localization strategy is adopted.The fine localization is performed by using the weighted radial projection method based on the coarse iris localization.According to the characteristics that the negative gradient directions of the iris boundary pixels point to the center point,the gradient radial projection value is selected as the screening basis for boundary parameter of the iris.At the same time,the boundary gray-scale change information is introduced as the weight to improve the accuracy of the method.In order to obtain an accurate description of the iris region of interest,this paper uses the adaptive gray-level threshold to detect the noises in the image.For the light spots,the threshold is set as a linear combination of the maximum and the average gray values of an effective region.As for the eyelash,the parameters of iris gray value Gaussian distribution are estimated by the divided region without eyelashes,and the threshold of eyelash detection is determined according to the statistical characteristics of Gaussian distribution.The algorithm proposed in this paper is tested on CASIA-Iris-Thousand database,and the segmentation accuracy of the algorithm is 99.1%.Experimental results show that this algorithm could effectively reduce the effect of noises in the image and has good robustness and fast calculation speed.