Research Of Iris Localization Algorithm Based On Multi-method Fusion

Iris localization is a key step in the iris-based identity authentication system,which directly affects the robustness and accuracy of the system.In order to improve the accuracy and robustness of iris localization,in this paper,a multi-method fusion iris localization algorithm is proposed,based on the quality evaluation of iris localization.And the iris image inpainting network is proposed in order to reduce the adverse effect of speckle noise on iris localization.In order to improve the accuracy and robustness of iris localization,this paper combines the integro-different operator and Hough transform method based on the quality evaluation of iris localization,and image inpainting and quality evaluation confidence judgment process are adopted to assist.First,under the premise that the inner and outer edges of the iris are circular,the the integro-different operator method is used to estimate the center and radius of the inner and outer edges of the iris.Then,the quality of the localization result will be judged,and the substandard samples are relocalization with Hough transform method.If the result of the relocalization still fails,the confidence of quality evaluation is reassessed,that is,the IOU of the two localization results is calculated.If the IOU of outer circle is greater than 0.75 and the inner circle is greater than 0.85,then the result of the quality evaluation is considered to be unreliable,and the localization result of integro-different operator will be retained.If the IOU is less than the threshold,then the image is reconstructed.The reconstruction process mainly aims at reducing the impact of speckle noise on localization.Taking the most advanced neural network as reference,the code-encode model is built for the current task.In order to enhance the effect of reconstruction,this study introduces the threshold-based speckle detection method and generates noise mask,assisting network to locate the position of noise.The reconstructed image will be segmented again by the calculus operator as the final result.It is also worth mentioning that when the reconstructed result PSNR and SSIM index is less than the threshold value,it is considered as the reconstruction failure,and the algorithm still retains the first localization result.In this study,a simulation experiment was conducted on the open database of Casia-Iris-Thousand provided by the Chinese academy of sciences.The localization accuracy of the proposed algorithm is improved from 92.68%to 95.08%using the integro-different operator method alone.The experimental results effectively proved the superiority of the proposed algorithm.