基于双路径残差网络的虹膜图像超分辨率重建

Super-resolution Reconstruction of Iris Image Based on Dual Channel Residual Network

低分辨率虹膜图像所含有效信息较少,实际应用到虹膜识别中会影响识别精度,而图像超分辨率重建技术能够有效解决这一问题。针对虹膜图像的结构和纹理差异,提出了适用于虹膜图像超分辨率的双路径网络结构,设计了双支路残差密集块提取深层虹膜特征,并采用后置放大策略重建图像。针对CASIA IrisV4虹膜图像库进行了实验,并与主流重建算法进行了比较,发现重建的图像结构清晰,纹理细节丰富。虹膜匹配实验结果表明,该算法的等错误率均为最优,重建图像具有良好的识别效果。

The low-resolution iris images contain less effective information,and their actual application will affect the recognition accuracy.Image super-resolution reconstruction technology can solve this problem effectively.Based on the difference between structure and texture of iris images,this paper proposes a dual-channel network structure which is suitable for iris images.We design dual-path residual dense blocks to extract deep features and use post-magnification strategy to reconstruct the final image.Experiments are conducted on the CASIA IrisV4 image library.Comparied with classic methods,the reconstructed images have clearer structure and richer texture details.The result of the iris matching experiments shows the EER are all optimal,which means the reconstructed iris images reach a better recognition performance.