基于小波变换的图像压缩感知深度重构网络
Compressed sensing deep reconstruction network based on wavelet transform
投稿时间:2020-02-19  修订日期:2020-02-19
DOI:
中文关键词: 深度学习  压缩感知恢复  小波变换  自适应采样
英文关键词: deep learning  compressed sensing  wavelet transform  adaptive sampling
基金项目:国家自然科学基金项目(面上项目,重点项目,重大项目)
作者单位E-mail
熊承义 中南民族大学 电子信息工程学院 xiongcy@mail.scuec.edu.cn 
李治邦 中南民族大学 电子信息工程学院 2017110192@mail.scuec.edu.cn 
高志荣 中南民族大学 计算机科学学院  
金鑫 中南民族大学 电子信息工程学院  
秦鹏飞 中南民族大学 电子信息工程学院  
摘要点击次数: 108
全文下载次数: 0
中文摘要:
      基于深度学习的压缩感知图像重构在当前得到了广泛关注。为了利用图像的先验特性改进压缩感知图像的重构质量,提出了一种基于小波变换的图像压缩感知深度重构方法。具体来说,基于迭代展开网络框架,设计的深度压缩感知重构网络包括采用梯度下降算法的图像冗余更新模块和采用小波变换的图像去噪模块,冗余更新模块和去噪模块交替多级级联。图像去噪模块保留小波低频分量不变,只对高频成分进行滤波去噪处理。提出的网络引入了自适应采样,以提高图像的采样效率。大量实验结果比较验证了所提方法的有效性.
英文摘要:
      Compressed sensing image reconstruction based on deep learning has received extensive attentions. To improve the reconstruction quality of compressed sensing images by utilizing the prior characteristics of images, a deep reconstruction network for compressed sensing images reconstruction based on wavelet transform is proposed in this paper. Specifically, based on the iterative unfolding network framework, the designed deep compressed sensing reconstruction network includs image redundancy update module using gradient descent algorithm and image denoising module using wavelet transform. The redundancy update module and denoising module are connected in an alternative multi-level cascaded fashion. The image denoising module keeps the low frequency components of the wavelet unchanged, and only performs filtering and denoising for the high frequency components. The proposed network introduces adaptive sampling to improve the efficiency of image sampling. Experimental results verify the effectiveness of the proposed method.
View Fulltext   查看/发表评论  下载PDF阅读器
关闭