【摘要】 为了提升指纹图像的分辨率,提出一种基于生成对抗网络(generation adversarial network,GAN)的指纹图像超分辨率重建方法。首先,采用高-低分辨率图像特征对训练生成神经网络,实现从低分辨率图像到高分辨率图像的映射学习;其次,为了解决指纹图像分辨率低、细节提取不足的问题,设计了多尺度递归网络作为生成网络,通过不同尺度的卷积层来进行特征提取,使生成的指纹图像更为清晰;最终,设计了一个新的损失函数,不断优化网络,指导生成高质量的指纹图像。实验结果表明,与对比方法相比,该方法在各指标上均有显著提升,并取得了较好的重建效果。更多还原

【Abstract】 To improve the resolution of fingerprint image, a method of super-resolution reconstruction of fingerprint image based on generation adversarial network(GAN) is proposed. Firstly, this method applied the high-low resolution image features to train the generated neural network to realize the mapping learning from the low-resolution image to the high-resolution image; Secondly, to solve the problem of low-resolution and insufficient detail extraction of fingerprint images, a multi-scale recursive network was designed as the generating network. The feature was extracted by convolution layers of different scales to make the fingerprint image clearer; Finally, a new loss function to optimize the network and guide the generation of high-quality fingerprint image was designed. The experimental results show that, compared with the comparative method, the proposed method was significantly improved in all indicators, and has achieved better reconstruction results.更多还原