Research On Super-resolution Reconstruction Algorithm Of Iris Image Based On Deep Learning

Iris recognition technology is an advanced human biometric technology,which uses the features contained in the iris area of the human eye as the key information to determine the identity of people.The iris feature has good uniqueness,stability,non-contact and anti-counterfeiting advantages,which is of great development potential,and has received extensive attention from people today.Image super-resolution reconstruction technology is the use of lowresolution images to restore high-resolution images,while zooming in the image,as much as possible to ensure the reconstruction accuracy and fidelity.This technology is widely used in medical,satellite,biometrics and other fields.In some uncontrollable application scenarios,such as surveillance and mobile biometrics,when the recognition person is far away from the iris collection device,the collected iris image lacks pixel resolution,and the iris image size is small,resulting in iris positioning errors and serious impacts identify performance.In order to solve this problem,this paper uses deep learning-based image super-resolution reconstruction technology to restore the low-resolution human eye image so that it can be used for iris recognition.The specific work is as follows:(1)Research on super-resolution reconstruction algorithm of iris images.Aiming at the low resolution of iris images collected at a long distance,which affects the effect of iris recognition,this paper proposes an adaptive weighted residual network suitable for iris images.The whole algorithm is based on the convolutional neural network,and fully takes into account the texture characteristics of the iris area,designing weighted residual dense blocks to extract rich iris features,and applying weighted residual connections to the entire network.In addition,in order to reduce the introduction of error information,the low-resolution iris image is directly used as the input of the network,and the extracted iris feature is mapped to the target size at the end of the network.The experimental results of CASIA-Iris V4-Lamp iris database show that the reconstructed iris image has clear and rich texture,which has outstanding advantages compared with other classic super-resolution reconstruction algorithms.(2)Design of iris image super-resolution analysis method.In addition to the subjective(visual)and objective(PSNR,SSIM)evaluation criteria for evaluating the quality of the reconstructed iris images,this paper also designs an iris recognition matching experiment to verify the effectiveness of the reconstructed iris image and the accuracy of recognition,reflecting the excellent performance of the adaptive weighted residual network.Iris segmentation adopts CINET method and Viterbi algorithm,and Iris recognition matching uses two-dimensional Gabor phase encoding algorithm and Hamming distance to simulate a real iris recognition scene.The ROC curve,equal error rate and intra-class and inter-class distribution,which are commonly used in iris recognition,are used to evaluate the quality of the reconstructed iris image.The experimental results show that the super-resolution iris image reconstructed by the adaptive weighted residual network is of high quality,perfectly reconstructs the texture features of the iris region,and the recognition effect is great,achieving the purpose of super-resolution reconstruction of the iris image.