一种Enlighten-GAN网络的指纹超分辨率重建方法北大核心CSCD

将低分辨率(LR)图像重建为高分辨率(HR)图像的主流模型是生成对抗网络(GAN)。然而,由于基于GAN的方法利用从其他图像中学习到的内容来恢复高频信息,在处理新的图像时往往会产生伪影。由于,指纹图像的特征比自然图像更加复杂。因此,将以前的网络应用于指纹图像,尤其是中等分辨率的图像,会导致收敛不稳定伪影效果更加严重。针对以上弊端,本文提出了一种Enlighten-GAN超分辨率方法,来解决指纹图像的重建问题。具体来说,我们设计了启发块来控制网络收敛到一个可靠的点,并利用自我监督分层感知损失以改进损失函数提升网络性能。实验结果证明Enlighten-GAN方法在指纹图像的重建效果性能上具有更加卓越的效果。

Enlighten GAN for super resolution reconstruction in mid resolution fingerprint images

The mainstream model for converting low resolution(LR)images into high resolution(HR)images is generating countermeasure network(GAN).However,because GAN based methods use the content learned from other images to recover high frequency information,artifacts often occur when processing new images.The characteristics of fingerprint image are more complex than natural image.Therefore,applying the previous network to fingerprint images,especially medium resolution images,will lead to unstable convergence and more serious artifacts.In view of the above disadvantages,this paper proposes an enlighten GAN super resolution method to solve the problem of fingerprint image reconstruction.Specifically,we design heuristic blocks to control the network convergence to a reliable point,and use self monitoring hierarchical loss perception to improve the loss function and improve the network performance.The experimental results show that Enlightens GAN method has better performance in fingerprint image reconstruction.