基于生成对抗网络的压缩感知图像重构方法

Compressed Sensing Image Reconstruction Method Based on Generative Adversarial Network

目的为了解决传统压缩感知图像重构方法存在的重构时间长、重构图像质量不高等问题,提出一种基于生成对抗网络的压缩感知图像重构方法。方法基于生成对抗网络思想设计一种由具有稀疏采样功能的鉴别器和具有图像重构功能的生成器组成的深度学习网络模型,利用对抗损失和重构损失2个部分组成的新的损失函数对网络参数进行优化,完成图像压缩重构过程。结果实验表明,文中方法在12.5%的低采样率下重构时间为0.009s,相较于常用的OMP算法、CoSaMP算法、SP算法和IRLS算法,其峰值信噪比(PSNR)提高了10~12 dB。结论文中设计的方法应用于图像重构时重构时间短,在低采样率下仍能获得高质量的重构效果。

The work aims to propose a reconstruction method of compressed sensing image based on generative adversarial network,in order to solve the problems of long reconstruction time and low quality of reconstructed image by traditional compressed sensing image reconstruction method.Based on the idea of generative adversarial network,a deep learning network model composed of discriminator with sparse sampling function and generator with image reconstruction function was designed.The new loss function composed of adversarial loss and reconstruction loss was used to optimize the network parameters and complete the process of image compression and reconstruction.Experiments showed that the reconstruction time required by the proposed method was 0.009 s at a low sampling rate of 12.5%,and the Peak Signal to Noise Ratio(PSNR)was 10-12 dB higher than that of the commonly used OMP algorithm,CoSaMP algorithm,SP algorithm and IRLS algorithm.When applied to image reconstruction,the proposed method require less reconstruction time and can still achieve a high-quality reconstruction effect at a low sampling rate.